Go deeper
Geopolitics and macroeconomics for forecasters
The domain knowledge a forecaster needs: geography and chokepoints, the balance of power, deterrence, alliances, sanctions and energy, then the accounting identities, inflation, the central bank, the yield curve, public debt, the exchange rate, financial crises and the dollar system — every lesson with a worked example, the reference class, the indicators to watch and a quiz.
Course outline
Read a chapter's lessons, take the chapter test, and finish with the exam simulation.
- Chapter 1The map and powerlessons 1–6Chapter test · 18 questions →
- Chapter 2Economic instruments and domestic politicslessons 7–12Chapter test · 18 questions →
- Chapter 3The mechanics of the macroeconomylessons 13–18Chapter test · 18 questions →
- Chapter 4Crises and the world systemlessons 19–22Chapter test · 12 questions →
- Chapter 5The forecaster's synthesislessons 23–24Chapter test · 6 questions →
The reading track
Chapter 1The map and power
- 01
Geography does not decide alone, but it fixes the costs: terrain, rivers and coasts make some options cheap and others ruinous, and this repeats more slowly than governments change.
Democratic Ideals and Reality · Halford J. Mackinder · 1919
Classical geopolitics starts from a simple idea: maps change more slowly than governments. In 1904 the British geographer Halford Mackinder read «The Geographical Pivot of History» to the Royal Geographical Society, calling the interior of Eurasia, out of reach of fleets, the ‘pivot area’. In «Democratic Ideals and Reality» (1919) he renamed it the Heartland and summed up the theory in the formula quoted above. Nicholas Spykman, in «America’s Strategy in World Politics» (1942) and «The Geography of the Peace» (1944), shifted the emphasis: decisive power lies in the Rimland, the populous and wealthy edge of Eurasia, not in its interior. The mechanism is constraint, not fate. Terrain, rivers, coasts and distances make some options cheap and others very expensive. A country with no natural barrier towards a strong neighbour invests permanently in defence or alliances; one with access to open seas has other commercial interests. The classic example is the East European plain, open from east and west alike: Napoleon’s army crossed it in 1812, the German army in 1941, and each time distance and winter worked against the invader. The limits are serious. Geographical determinism explains too much and too little: the same geography accompanied both the Russian empire and its collapse. Aircraft, missiles and satellites have compressed the distances Mackinder thought decisive, and the borders of the ‘Heartland’ have always been disputable. For the forecaster, geography is a structural reference class: it says what repeats (invasion routes, forced crossing points), not when. You use it to rule out scenarios that demand the logistically impossible, not to put a date on any scenario.
Who rules East Europe commands the Heartland; who rules the Heartland commands the World-Island; who rules the World-Island commands the world.Halford J. Mackinder, «Democratic Ideals and Reality: A Study in the Politics of Reconstruction», Constable, 1919, cap. IV «The Freedom of Nations» (traducerea noastră)
Why it matters The Black Sea, the Danube and the Carpathians shaped the routes of armies and goods in Romania before any government did, and the Danube–Black Sea Canal (1984) cut roughly 400 km from the trip between the Danube and Constanța.
In full →Collapse ↑
Worked example: Romania between the Carpathians, the Danube and the Sea. On the map the country is shaped like an arc: the Eastern, Southern and Western Carpathians enclose Transylvania like a fortress, while outside them, to the east and south, lies an amphitheatre of plains and hills open to the Black Sea. The Danube, to the south, is at once a long border and a trade route, and at its mouths lies the Delta, where the river meets the sea. Three consequences follow directly from the terrain.
The first is the passage between Moldavia and Wallachia. Between the last spurs of the Curvature Carpathians and the Danube there remains a narrow corridor, the Focșani Gate, which armies have crossed over the centuries. In the summer of 1917 the battle of Mărășești was fought on this very corridor: the Romanian army, reorganised after losing Bucharest in December 1916, stopped the German offensive there. The geographical advantage was real but not sufficient: the front held because the defenders had artillery, reserves and command, not terrain alone.
The second is the Danube as a road. The Danube–Black Sea Canal, opened in 1984, shortens the route to Constanța by about 400 km compared with the detour through the Delta. Access to the Black Sea then ties the country to the Bosporus and Dardanelles, controlled by Turkey under the Montreux Convention (1936), the subject of the next lesson.
The third is the position relative to the Rimland. Romania lies on the western edge of the Pontic steppe, exactly where Mackinder saw ‘East Europe’. No fate can be assigned from a map, but it gives a permanent cost of neighbourhood and a permanent gain from trade routes. The analyst’s next step is to turn these constraints into classes of observable events: tonnage through Constanța, traffic on the Danube, incidents in the Black Sea.
For the forecaster: there is no base rate for ‘geography’ in general; the reference class is built on a concrete, measurable object. For the example above the natural class is the annual series of cargo traffic of the ports of Constanța, Midia and Mangalia, published by the Constanța Maritime Ports Administration: you count in how many of the last ten years traffic rose over the previous year and what explained each fall (drought, war, grain prices).
Indicators to follow, with the source of each: annual and monthly tonnage of the maritime ports (Constanța Maritime Ports Administration); the Danube level at downstream stations (the National Institute of Hydrology and Water Management); Ukrainian grain exports along the Danube and across the Black Sea (customs data and reports from Ukraine’s infrastructure ministry); navigation incidents in the Black Sea (notices from the International Maritime Organization and the Turkish Navy).
The typical trap is to turn a constraint into a prediction: ‘the route is forced, so traffic will grow’. The constraint says the route exists, not how much moves along it; volume depends on prices, harvests, sanctions and war. A second trap: choosing the threshold from the latest year when that year was abnormal.
Example scorable question: ‘Will total cargo traffic of the ports of Constanța, Midia and Mangalia in calendar year 2027 exceed the 2026 figure, according to data published by the Constanța Maritime Ports Administration by 31 March 2028?’ It resolves yes/no on the official figures; if the administration does not publish the aggregate total, the question is voided.
- 02
The sea is a common, but traffic runs through a few narrow lanes; whoever can close or defend such a lane holds leverage out of proportion to its size.
The Influence of Sea Power upon History, 1660–1783 · Alfred Thayer Mahan · 1890
Alfred Thayer Mahan, an American naval officer, published «The Influence of Sea Power upon History, 1660–1783» in 1890. His thesis: the history of great powers is largely explained by control of commercial sea routes, and such control requires a battle fleet, bases and a trade worth defending. He lists the conditions of sea power: geographical position, physical conformation of the coasts, extent of territory, population, national character and the character of the government. The book influenced naval policy in the United States, Germany and Japan in the following decades. The modern mechanism is the chokepoint: a narrow passage through which a large share of a flow moves. According to the US Energy Information Administration (EIA), about 20 million barrels a day passed through the Strait of Hormuz in 2024, roughly one fifth of world consumption of petroleum liquids. The Houthi attacks in the Red Sea, begun in November 2023, cut Suez transits by 42% in only two months, according to UNCTAD, and ships went round Africa, adding days to Asia–Europe routes. The Bosporus is a legal case: the Montreux Convention (1936) gives Turkey control over the passage of warships, above all in wartime. Limits: sea power does not automatically mean the will to use it; alternative routes, stocks and pipelines blunt the shock; and cheap missiles and drones have lowered the cost of threatening a lane. For the forecaster the useful question is not ‘can it be closed?’ but ‘for how long, with what detour and with what response?’.
The first and most obvious light in which the sea presents itself from the political and social point of view is that of a great highway; or better, perhaps, of a wide common, over which men may pass in all directions, but on which some well-worn paths show that controlling reasons have led them to choose certain lines of travel rather than others.Alfred Thayer Mahan, «The Influence of Sea Power upon History, 1660–1783», Little, Brown, 1890, cap. I «Discussion of the Elements of Sea Power» (traducerea noastră)
Why it matters Hormuz carries about a fifth of world oil consumption, yet the pipelines of Saudi Arabia and the United Arab Emirates can bypass at most about 2.6 million barrels a day, or 13% of the flow.
In full →Collapse ↑
Worked example: ‘Ever Given’, 23–29 March 2021, and the arithmetic of Hormuz. On 23 March 2021 the container ship ‘Ever Given’, about 400 m long, wedged itself across the Suez Canal, at a single-lane stretch. The canal stayed blocked until 29 March: six days in which hundreds of ships waited at either end or chose the detour round Africa. For the analyst the episode tests a simple claim: ‘Suez is a chokepoint’.
What it shows: an accident with no political intent could halt a significant part of Asia–Europe trade; but the blockage was short and the queue cleared within weeks. The conclusion is neither ‘lanes are fragile’ nor ‘they are safe’, but that duration matters as much as the existence of a blockage. For the Red Sea the figures after November 2023 show something else: not an accident but a persistent threat, and UNCTAD measured a 42% fall in transits within two months; the routes round Africa add, according to shipping lines, some 10–14 days on Asia–Europe voyages.
Now the arithmetic of Hormuz, with EIA figures for 2024: about 20 million barrels a day pass through the strait; the pipelines of Saudi Arabia and the Emirates can bypass, by EIA’s estimate, about 2.6 million barrels a day. The ratio: 2.6 divided by 20 is 0.13, so 13% of the flow has a bypass. That leaves 20 − 2.6 = 17.4 million barrels a day with none, or 87%. Over a year, 17.4 million × 365 days is 6,351 million barrels, about 6.4 billion.
These numbers do not predict a closure; they fix the size of the shock if one occurred, plus one important detail: over 80% of the crude passing through Hormuz goes to Asian markets, so the pressure is not equal for all consumers.
For the forecaster: the base rate of ‘lane blockages’ is not one figure but several classes. The first: accidental blockages (like ‘Ever Given’), lasting days; the second: military or quasi-military closures, rarer, of unknown duration in advance; the third: disruption by threat to ships (the Red Sea), which can last years. You build each class by counting episodes in the trade press and in shipping-line reports, then computing the median duration.
Indicators to follow, with the source: daily transits through Suez and Bab el-Mandeb (the International Monetary Fund’s PortWatch portal); oil flows and bypass routes (the US Energy Information Administration); container freight rates (the Drewry World Container Index); notices to mariners (the International Maritime Organization and the EU naval operation Aspides).
The typical trap is to confuse traffic volume with importance: a high-traffic strait may have a cheap bypass, while a low-traffic one may be irreplaceable (access to a single port, for instance). A second: forecasting ‘closure’ without defining it (zero transits? a 50% fall? a ban on one flag?).
Example scorable question: ‘According to PortWatch, will the average daily number of transits through the Suez Canal in December 2027, computed as the mean of all days of the month, be below 50, on data published by 15 January 2028?’ It resolves yes/no from the published series; if the series is no longer available, the question is voided.
- 03
States arm out of fear, and one’s arming frightens another; so the balance of power does not maintain itself, and neighbours react to threat more than to mere power.
History of the Peloponnesian War · Tucidide (trad. Richard Crawley) · 1874
In the Melian dialogue (416 BC, in book five), Thucydides has the Athenians say that right matters only between equals. The sentence became the emblem of realism in international relations. Kenneth Waltz gave the theory a strict form in «Theory of International Politics» (1979): the international system is anarchic, with no authority above states, and states, whatever their regime, behave alike because the structure forces them to fend for themselves. Robert Jervis showed in «Cooperation under the Security Dilemma» (World Politics, 1978) how this structure produces a trap; the term had been launched by John Herz in 1950. The mechanism of the security dilemma: what a state does to feel safe (arms, alliances, bases) makes its neighbours less safe, and they respond in kind. Jervis shows that the intensity depends on two things: whether attack is cheaper than defence and whether defensive weapons can be told apart from offensive ones. Stephen Walt added in «The Origins of Alliances» (1987) that states balance first against threat (power, geographical proximity, offensive capability, perceived intentions), not against the greatest power. Limits: the model assumes unitary, rational states; sometimes states align with the strong (bandwagoning), and domestic politics can break the logic. For the forecaster the useful question is whether a neighbour reads a military move as offensive or defensive, since the reaction depends on the reading.
...since you know as well as we do that right, as the world goes, is only in question between equals in power, while the strong do what they can and the weak suffer what they must.Tucidide, «Istoria războiului peloponesiac», V.89, dialogul melian, traducerea engleză a lui Richard Crawley (1874) (redarea română: traducerea noastră)
Why it matters At the Hague summit in June 2025 NATO allies agreed to reach 5% of GDP on defence and security by 2035, a jump the security-dilemma literature reads as a response to a perceived threat and, at the same time, as a signal the neighbours will read in their turn.
In full →Collapse ↑
Worked example: the Anglo-German naval race and London’s choice. Britain based its security on its fleet: from the Naval Defence Act (1889) it aimed for a navy at least equal to the next two naval powers combined. Germany, after Tirpitz’s naval laws of 1898 and 1900, began building a battle fleet, which it justified as a ‘risk’ Britain should not take in a war with it. From Berlin’s view it was a fleet of prestige and bargaining; from London’s, a fleet concentrated in the North Sea, a few days from the English coast.
That is how the security dilemma works. London’s answer was HMS Dreadnought, launched in February 1906, a ship that made existing fleets obsolete and reduced the British numerical lead: now only new-type ships counted, and Germany could build them. Germany responded by building too. Attempts at understanding, among them Haldane’s mission to Berlin in 1912, failed, because Berlin wanted a British neutrality clause that London could not give.
What did London do diplomatically? It did not ally with the strongest and did not bandwagon with Germany. It concluded the Entente Cordiale with France in 1904 and a convention with Russia in 1907, though it had had years of colonial rivalry with both. Walt would say: Britain balanced against the state that threatened it most directly (the German fleet, its closeness to the coast, the pace of building), not against the state with the largest land army.
The lesson for the analyst: an arms spiral does not require aggressive intentions on either side; it requires only that each read the other’s posture as offensive.
For the forecaster: the security dilemma has no base rate of its own, but it has observable reference classes. The first: pairs of states with simultaneously rising military spending, measured by SIPRI; you count how many such pairs in the last 30 years ended in armed conflict and how many concluded a limitation agreement (such as arms-control treaties). The second: episodes in which an alliance took in new members and the neighbour responded militarily.
Indicators, with the source: states’ military expenditure (the SIPRI database); announced exercises and deployments (communiqués of defence ministries and of NATO); arms-control treaties in force (the UN Office for Disarmament Affairs); official rhetoric, quoted from public speeches.
The typical trap: treating any rise in a defence budget as a sign of aggression. It may answer a real threat, and not answering may be more dangerous. A second trap: assuming the other side sees things as you do; the reading of a posture depends on its perception, not on your intent.
Example scorable question: ‘Will at least 20 NATO member states exceed 2% of GDP in military spending, according to the annual expenditure report published by NATO in 2027 (data for 2026), published by 31 December 2027?’ It resolves by counting from the official table; if NATO does not publish the annual report, the question is voided.
- 04
Deterrence prevents an action not yet begun, compellence forces a stop or a withdrawal; both rest on credibility, meaning what words cost if they are not followed by deeds.
Arms and Influence · Thomas C. Schelling · 1966
Thomas Schelling, an economist and game theorist, published «Arms and Influence» in 1966. The basic idea: military power is used not only when fighting but also when threatening; the capacity to do harm is a bargaining currency. Schelling distinguished deterrence, which persuades an adversary not to begin an action, from compellence, which persuades him to stop or undo something already begun. The first needs a threat that stays in reserve; the second needs a deadline, otherwise the adversary can delay indefinitely. The mechanism is credibility. A threat is credible if the adversary believes you will carry it out even after carrying it out costs you. Schelling describes ways of making it credible through commitments that tie your hands: troops left as a ‘tripwire’, a public deadline, a reputation put at stake. In 1962 the Cuban missile crisis showed both: on 22 October the US chose a ‘quarantine’ of Cuba, less escalatory than an attack, and on 28 October Khrushchev announced the withdrawal of the missiles. Limits: credibility is judged by the one threatened, not by the one threatening; cheap signals do not build it, and a threat that is too large can make acceptance politically impossible for the other side. For the forecaster the useful question is what the signal costs the one who sends it: words are cheap, mobilisation and decisions with high domestic cost are not.
The power to hurt is bargaining power. To exploit it is diplomacy—vicious diplomacy, but diplomacy.Thomas C. Schelling, «Arms and Influence», Yale University Press, 1966, cap. 1 «The Diplomacy of Violence» (traducerea noastră)
Why it matters On 28 October 1962, after 13 days of crisis, the Soviet missiles in Cuba were withdrawn, and the American Jupiter missiles in Turkey were removed in April 1963, as part of an understanding not public at the time.
In full →Collapse ↑
Worked example: the Cuban missile crisis, 16–28 October 1962, read with the two concepts. On 16 October 1962 President Kennedy learned that photographs from a U-2 aircraft showed launch sites for Soviet medium-range missiles in Cuba. For a week a small group of advisers weighed options, from an air strike to negotiation. On 22 October Kennedy announced on television a ‘quarantine’ of Cuba, enforced from 24 October, and demanded the withdrawal of the missiles.
This is compellence in Schelling’s sense: a demand to undo a situation already created, not to avoid a future one. The quarantine, rather than an attack, was chosen because it left Moscow room to manoeuvre: a Soviet ship could turn back without anyone firing. It is exactly the form of pressure Schelling calls ‘the power to hurt held in reserve’. On 27 October a Soviet missile shot down an American U-2 over Cuba, and the pilot, Major Rudolf Anderson, died; Washington chose not to respond at once with force. On 28 October Khrushchev announced on radio the withdrawal of the missiles.
What the chronology shows: a public deadline (the quarantine of 24 October), a cost to the sender (Kennedy had publicly staked his prestige) and an honourable exit for the adversary. It was later learned that there was also a non-public understanding on the American Jupiter missiles in Turkey, withdrawn in April 1963. They show what a successful compellence may require: a way out for both sides.
The lesson for the analyst: when you read a threat, ask what the adversary is asked to do (stop or refrain), whether there is a deadline, and what exit he is left.
For the forecaster: a reference class for crises with reciprocal threats exists and is documented: the International Crisis Behavior (ICB) database, which codes international crises from 1918 onward, with outcomes (agreement, stalemate, victory). A base rate is built on it by selecting crises with an explicit threat and counting in how many violence followed; the analyst takes the exact figure from the current version of the database, not from memory.
Indicators, with the source: official statements with a deadline and a condition (communiqués of governments, of NATO, of the UN Security Council); visible mobilisations and deployments (defence ministry reports and commercial satellite imagery published by research institutes); costly diplomatic moves (recalling ambassadors); missile tests (the CSIS Missile Threat database).
The typical trap: taking words as a signal. A harsh statement may be aimed at the domestic audience, not the adversary; what counts is what the sender has already done, not what he said he would do. A second trap: assuming maximum pressure leads to concession; a threat that is too large leaves the adversary only the option of conceding in public, often impossible.
Example scorable question: ‘Will North Korea carry out, between 1 January and 31 December 2027, at least one intercontinental ballistic missile test, according to the CSIS Missile Threat database consulted on 31 January 2028?’ It resolves yes/no from the database; if it is no longer updated, the list published by Japan’s Ministry of Defense is used, and if that does not exist either, the question is voided.
- 05
War is a joint loss, so a deal should always exist; it occurs when private information, the lack of a credible commitment or an indivisible stake make the deal unreachable.
Rationalist Explanations for War · James D. Fearon · 1995
In «Rationalist Explanations for War» (International Organization, 1995), James Fearon started from an awkward question: war destroys resources, so both sides should prefer a negotiated settlement. If they fight nonetheless, something prevents a deal. Fearon identifies three rational explanations: private information with incentives to misrepresent it, commitment problems and the indivisibility of the stake. The other popular explanations (irrationality, mistakes, domestic politics) are not rejected, but these three can be modelled without assuming leaders do not think. The mechanism. Private information: each side knows its own capability and resolve better than the adversary does, and each has an interest in seeming stronger; so statements are not trustworthy, and fighting sometimes becomes the way of finding out. The commitment problem: a deal made today does not hold if one side will grow stronger tomorrow, so the other prefers to strike first. Indivisibility: some stakes (a holy city, a throne) cannot be shared, though Fearon notes that compensation can usually be negotiated. Limits: the model assumes unitary, rational actors, and for forecasting the question remains how you measure ‘private information’. What it means for the forecaster: wars are rare, and most disputes end without fighting, so the base rate is low; it rises when signals appear that remove uncertainty (costly mobilisation) or when a rapid shift of power makes a commitment no longer credible.
Because war is costly and risky, rational states should have incentives to locate negotiated settlements that all would prefer to the gamble of war.James D. Fearon, «Rationalist Explanations for War», International Organization, vol. 49, nr. 3, 1995, pp. 379–414 (formularea din corpul articolului, redată din memorie; traducerea noastră)
Why it matters Russia’s attack on Ukraine of 24 February 2022 followed a troop build-up in spring 2021 that ended without fighting, so the same signal was a false alarm once and correct the second time.
In full →Collapse ↑
Worked example: two years of signals before 24 February 2022, and the arithmetic of a base rate. The Correlates of War (COW) list, version 4.0 for 1816–2007, counts 95 interstate wars, defined by at least 1,000 battle deaths. The calculation: the period 1816–2007 has 2007 − 1816 + 1 = 192 years; 95 divided by 192 gives about 0.49 wars a year, so on average one every just over two years across the whole international system. For a specific pair of states the annual rate is incomparably smaller: these 95 wars are spread across thousands of dyad-years.
Now the case. In spring 2021 Russia massed troops near Ukraine, then withdrew some of them in April, without fighting. The alarm proved false. In autumn 2021 the build-up resumed, larger, and lasted for months; on 24 February 2022 the invasion began. From Fearon’s perspective the difference between the two episodes is not the number of troops but the information they revealed. A cheap signal (a deployment that can be cancelled with no political loss) does not remove uncertainty; a costly one (months of visible preparation, maximalist public demands) does, because the sender loses much if he backs down.
What the case shows a forecaster: there is a low base rate (interstate war is rare), but costly signals update that rate strongly. After the 2021 build-up many analysts put small probabilities; after months of visible preparation a few raised them sharply. No single signal decided; it was the accumulation of those that cannot be withdrawn for free.
The lesson: ‘mobilisation at the border’ does not mean invasion, but it is not zero either. The right question is what the signal costs, not how large it is.
For the forecaster: a reference class for ‘the outbreak of an interstate war’ exists: Correlates of War (the ‘Inter-State Wars’ list, 1816–2007) with 95 wars and a threshold of 1,000 battle deaths, complemented by the UCDP/PRIO series, with a threshold of 25 deaths a year. The two thresholds give different base rates, so the question must name the source and the threshold.
Indicators, with the source: the UCDP database (Uppsala University) for active armed conflicts; defence ministry reports and commercial satellite imagery for deployments; official statements with a deadline and maximalist demands; the pace of arming (SIPRI). For each, you note whether the signal is cheap or costly.
The typical trap: confusing a dispute with a war. Most territorial disputes never led to fighting, and a base rate computed only on the cases that escalated is inflated and misleading. A second trap: taking every build-up as an invasion or as a bluff; both happen, and the difference lies in the cost.
Example scorable question: ‘Will the version published in 2028 of the UCDP/PRIO Armed Conflict dataset contain at least one interstate-type conflict active in calendar year 2027?’ It resolves yes/no from the dataset; if UCDP does not publish the version by 31 December 2028, the question is voided.
- 06
An alliance guarantee is worth as much as it is believed by allies and adversaries: it does not automatically promise war but the action each deems necessary, and between abandonment and entrapment there is no perfect dosage.
The North Atlantic Treaty (Washington, 1949), articolul 5 · Părțile Tratatului Atlanticului de Nord · 1949
Article 5 of the North Atlantic Treaty (Washington, 1949) is the core of NATO: an armed attack against one of the members in Europe or North America is considered an attack against them all. The text does not, however, promise an automatic response: each party will take ‘such action as it deems necessary, including the use of armed force’. The wording leaves open what measures each state takes, so that the decision to go to war stays with its own constitutional bodies. The clause has been invoked only once, after the attacks of 11 September 2001; the North Atlantic Council decided so on 12 September, on condition that the attack came from abroad. The mechanism of the lesson comes from alliance theory. Glenn Snyder, in «The Security Dilemma in Alliance Politics» (World Politics, 1984), shows that allies fear two things: abandonment (being left alone in need) and entrapment (being dragged into the partner’s conflict). The firmer the guarantee, the more the fear of abandonment falls and the fear of entrapment rises, and the reverse. That is why small states ask for clear guarantees while large ones keep ambiguity. Limits: an alliance is not a contract judged by a court; credibility is judged by capabilities, military presence and political will. For the forecaster the spending target is an indicator of commitment: at the Wales summit (2014) allies set 2% of GDP, and at The Hague (25 June 2025) they agreed 5% by 2035, of which 3.5% is core defence. The promise counts as much as its execution.
...each of them, in exercise of the right of individual or collective self-defence recognised by Article 51 of the Charter of the United Nations, will assist the Party or Parties so attacked by taking forthwith, individually and in concert with the other Parties, such action as it deems necessary, including the use of armed force, to restore and maintain the security of the North Atlantic area.Tratatul Atlanticului de Nord, Washington, 4 aprilie 1949, art. 5, text oficial NATO (redarea română: traducerea noastră)
Why it matters Article 5 has been invoked only once in over 75 years although NATO has had dozens of crises, so the base rate of invocation is very low, and the presence of troops matters more than the text.
In full →Collapse ↑
Worked example: Article 5 in September 2001 and the arithmetic of the spending target. On 11 September 2001 terrorist attacks struck New York and Washington. The next day, 12 September, the North Atlantic Council decided that, if it were confirmed that the attack had been directed from outside the United States, it would fall under Article 5. Confirmation came on 2 October, and measures were applied from October 2001: among others, NATO AWACS surveillance aircraft flew over the United States. A clause invoked only once allows no conclusions from its frequency; what is visible, however, is the form of the response: not automatic armed intervention but ‘such action as each deems necessary’.
Now the spending target. At the Wales summit (2014) allies committed to reach 2% of GDP on defence within a decade. At The Hague on 25 June 2025 they set a commitment of 5% of GDP by 2035: at least 3.5% for core defence and up to 1.5% for critical infrastructure, resilience and industry. Take a fictional country with a GDP of 300 billion euros. At 2% it spends 0.02 × 300 = 6 billion. At 3.5%: 0.035 × 300 = 10.5 billion, so 4.5 billion more. At 5%: 0.05 × 300 = 15 billion, which is 2.5 times the 2% level (15 divided by 6). If the economy grows 30% meanwhile, the same percentages mean larger sums: 5% of 390 billion is 19.5 billion.
What the calculation shows: a target expressed as a share of GDP does not say how much is spent, only how it moves with the economy. That is why the analyst follows two series: the share of GDP and the sum in the country’s currency at constant prices.
The lesson: a guarantee is assessed by what it costs now and what it costs at maturity.
For the forecaster: the base rate of invoking Article 5 is one case in almost 77 years (1949–2026), so about 1.3% a year; for a given year it is adjusted by concrete signals: armed attacks on an ally’s territory, consultations under Article 4 (which are more frequent and can be counted separately), mobilisations. A second class: how many states reach the spending target in a year, a public number.
Indicators, with the source: NATO’s annual report on defence expenditure (figures by country, as a share of GDP); the SIPRI military expenditure database; communiqués of the North Atlantic Council (Article 4 consultations are announced publicly); national budgets voted in parliaments.
The typical trap: confusing promise with execution. A summit commitment is not actual spending; between them lie budgets, contracts and industrial capacity. A second trap: taking the absence of invocation as proof of credibility or of its lack; Article 5 may deter precisely by not being invoked.
Example scorable question: ‘Will NATO report, in its report on defence expenditure published in 2027 (estimates for 2026), at least 25 member states with core defence spending of at least 2% of GDP?’ It resolves by counting the states in the official table; if NATO does not publish the report by 31 December 2027, the question is voided.
Chapter 2Economic instruments and domestic politics
- 07
Whoever controls a node through which the world's flows pass, from bank messaging to semiconductors, can see what circulates and cut others off: interdependence becomes an instrument of coercion.
Weaponized Interdependence: How Global Economic Networks Shape State Coercion · Henry Farrell și Abraham L. Newman · 2019
In 1945, in National Power and the Structure of Foreign Trade, Albert Hirschman showed that trade between a large country and a small one is not symmetric: the small one loses more if the relationship breaks, and the large partner can turn that asymmetry into influence. Henry Farrell and Abraham Newman moved the idea from pairs of states to networks. In Weaponized Interdependence (International Security, 2019) they argue that the global economy has central nodes, such as bank messaging, dollar clearing or semiconductor production. The state with jurisdiction over a node gains two powers: the panopticon effect, meaning it can see the flows that pass through it, and the chokepoint effect, meaning it can cut an adversary off. The mechanism needs an asymmetric network, jurisdiction over the node and domestic institutions willing to act. SWIFT, a Belgian messaging cooperative, disconnected Iranian banks targeted by European Union sanctions in March 2012. In February 2022 Western governments immobilised part of the Bank of Russia's reserves, about 300 billion dollars by their own estimates, out of a total of roughly 630 billion. A node is not eternal: every use shows the target how costly it is to depend on it. The limits lie exactly there. Alternatives (local payment systems, settlement in other currencies, reserves held in gold) raise the cost for whoever uses the weapon, and a node used too often loses its centrality. For the forecaster, the useful question is not ‘will state X use the weapon?’ but who has jurisdiction over the node, how quickly the target can replace it and what the issuer loses if it does. Capacity and decision are different things.
Increasingly, states are employing global economic networks to fulfill their strategic objectives.Henry Farrell și Abraham L. Newman, «Weaponized Interdependence: How Global Economic Networks Shape State Coercion», International Security, vol. 44, nr. 1, 2019, pp. 42–79, rezumatul articolului (traducerea noastră)
Why it matters The immobilisation of the Bank of Russia's reserves in 2022 showed that reserves held in other jurisdictions are not guaranteed, and the debate on where central banks keep their reserves was reopened.
In full →Collapse ↑
Worked example: from network to numbers. Take the Bank of Russia's reserves first. At the start of 2022 they stood at about 630 billion dollars, a figure given in the Australian Parliamentary Library note of March 2022. Western governments immobilised, by their own estimates, about 300 billion: 300 / 630 ≈ 0.476, almost half. The rest, held in gold, in yuan or in accounts in jurisdictions that did not join the coalition, stayed accessible. The chokepoint effect was therefore partial, and it hit exactly the part held in the coalition's currencies and banks.
The second calculation is Hirschman's: concentration. For a hypothetical country whose exports are split among four partners in the proportions 60%, 25%, 10% and 5%, the sum of the squared shares is 0.36 + 0.0625 + 0.01 + 0.0025 = 0.435. If the same exports were split equally, 4 × 0.25² = 0.25, the minimum possible for four partners. The higher the index, the more credibly the dominant partner can threaten a rupture. The figure does not say whether the threat will be carried out, only how much it would cost the target to bear it.
The third step is the comparison between 2012 and 2022. The node was the same, SWIFT, and jurisdiction over it stayed the same; what changed was the size of the target and the available alternatives. In between, in 2015, the Chinese CIPS system appeared, a partial alternative for settlement in yuan. A forecaster who had only counted how many disconnections had occurred could not have deduced from 2012 what happens with a much larger economy; he had to ask what alternatives to the node existed and how much of the flows passed through them. The base rate says how often the weapon is used, the structure of the network says what it costs.
For the forecaster: the reference class has no standard count. It is built from sanctions databases (for example the Global Sanctions Database), keeping only financial measures: disconnections from payment systems, reserve freezes, restrictions on access to the dollar. Each row is counted once, on the date of the legal act, not of the political announcement. Without such a list, any ‘rate’ is only an impression.
Indicators to watch, with the public source of each. The share of currencies in allocated reserves: the IMF, COFER survey, quarterly. The share of the yuan in international payments: SWIFT's monthly renminbi report. The activity of alternative payment systems: CIPS reports. New sanctions acts: the Official Journal of the European Union and US Treasury releases.
The typical trap: confusing capacity with decision. That a state can cut a node says nothing about when it will do so, and the cost to the issuer rises with every alternative that matures. The second trap: treating an announcement as an act. A political statement disconnects no bank; the date of entry into force is what counts.
A scorable example question, illustrative: ‘Between 1 January and 31 December 2027, will SWIFT withdraw access from at least one bank in a country whose banks were connected on 1 January 2027, according to a SWIFT statement or an act published in the Official Journal of the EU?’ Resolution source: SWIFT statements and the Official Journal; the date that counts is that of entry into force; a later suspension does not change the outcome.
- 08
Sanctions seem to succeed in about a third of cases if you count generously and in under 5% if you count strictly: the gap comes from the definition of success, not from the data, and it shows when they really work.
Why Economic Sanctions Do Not Work · Robert A. Pape · 1997
The first great count of sanctions came from Gary Hufbauer, Jeffrey Schott and Kimberly Elliott in Economic Sanctions Reconsidered (first edition 1985; the third edition, of 2007, with Barbara Oegg, covers 174 cases from the twentieth century). The often-cited result: ‘at least partial’ success in about a third of cases; in the 1990 edition, 40 successes out of 115 cases. In 1997, in Why Economic Sanctions Do Not Work (International Security), Robert Pape redid the list: of the 40 claimed successes, only 5 survive a strict definition. The difference between the two figures lies in the definition. HSE scores a partial or modest success as well; Pape requires that the main objective was achieved through economic pressure, not through the threat of force or other factors. The classic failure is Italy in 1935–1936: after the invasion of Ethiopia (3 October 1935), the League of Nations' sanctions took effect on 18 November 1935, but oil stayed outside them, and Italy completed the conquest in May 1936; the League's Assembly voted to lift them on 4 July 1936. Successful cases share a few traits: a limited aim, a target dependent on the issuer, a broad coalition and a target with no alternatives. The limits of the counts lie in selection and in the counterfactual: sanctions that were only threatened, and worked without being applied, rarely appear in the data, and what would have happened without sanctions is not observed. For the forecaster, the base rate depends on the aim: freeing a prisoner or winning a trade concession has different odds from changing a regime. The two figures do not contradict each other; they answer different questions.
The world would have to change considerably before sanctions could become a credible alternative to force.Concluzia lui Robert A. Pape, «Why Economic Sanctions Do Not Work», International Security, vol. 22, nr. 2, 1997, pp. 90–136, redată după rezumatul National Research Council (2000), nu verificată pe textul articolului (traducerea noastră)
Why it matters Sanctions packages are now counted in dozens, but their success depends on the aim: whoever asks ‘will they work?’ without saying ‘for what?’ has not yet formulated a question.
In full →Collapse ↑
Worked example: the same data, two definitions. Take a set of 115 sanctions cases from 1914–1990, in which the original authors claim 40 successes. Their rate: 40 / 115 ≈ 0.348, about 35% (in the 1990 edition the figure cited is 34%). Pape takes the 40 cases again and accepts 5. The fraction of the 40 claimed: 5 / 40 = 0.125, that is 12.5%. The fraction of the total of 115: 5 / 115 ≈ 0.043, under 5%. The same data, the same list of cases, so the difference from 35% to 4% comes entirely from the definition of success.
A case that shows the mechanism from the inside is Italy after the invasion of Ethiopia. Italy attacked on 3 October 1935 and the League of Nations' sanctions took effect on 18 November 1935, but the product Italy needed most for the war, oil, was not included, and the main suppliers were not part of the coalition. Italy completed the conquest of Ethiopia in May 1936, and the League's Assembly voted to lift the sanctions on 4 July 1936. In the terms of the lesson: a large aim (withdrawal from a country already occupied), a target that had alternatives, a coalition with gaps. The result is a failure from any perspective.
Now, how to use this in a forecast. Before putting a probability on ‘will the sanctions succeed?’, write down the aim: what is the objective, how many states apply the measure, how much of the target's trade passes through them and what alternatives the target has. If the answers are ‘large aim, partial coalition, alternatives available’, the relevant reference class is that of strictly successful sanctions, not the aggregate average. If the aim is modest, for example the release of some detainees, the reference class is different and the base rate is higher. The 35% figure is therefore not a probability but an average over incomparable aims.
For the forecaster: the reference class is chosen by aim. The sources of the counts are the academic databases: HSE (with the case set from the book published by the Peterson Institute for International Economics), the Global Sanctions Database and TIES (Threat and Imposition of Sanctions). Each has its own definition of success; before using a figure, read the definition and check whether the aim of your question is the same size as the aims of the cases counted. If it is not, say explicitly that the base rate is an approximation.
Indicators to watch, with the public source. New acts and lifting of measures: the Official Journal of the European Union and the SDN list of the US Treasury (OFAC). The target's trade with those applying the measures: Eurostat, UN Comtrade, the country's customs data. The target's exchange rate and reserves: the target's central bank and the IMF. A lagging but useful indicator: if the target redirects its trade toward states that do not apply the sanctions, the coalition has gaps.
The typical trap: taking the political statement as proof of success. A leader who ‘yields’ may do so for a reason unrelated to the sanctions, and a leader who does not yield proves nothing about them: if you do not know the counterfactual, you cannot attribute the outcome.
A scorable example question, illustrative: ‘Between 1 January and 31 December 2027, will OFAC remove from the SDN list at least 10 entries (persons or entities) associated with the same sanctions programme, according to OFAC's public recent-actions archive?’ Resolution: the OFAC archive; removed entries are counted, not amended ones; the date that counts is that of publication of the removal notice.
- 09
When energy demand cannot adjust quickly, a small supply shortfall requires a big price jump: short-run elasticity, not the size of the shock, decides how violently the market moves.
The Prize: The Epic Quest for Oil, Money and Power · Daniel Yergin · 1991
Daniel Yergin placed oil at the centre of twentieth-century political history in The Prize (1991), which received the Pulitzer Prize in 1992. The economic idea behind it is simple: in the short run energy has inelastic demand, because people do not change their car, power plant or heating from one day to the next. An economy's vulnerability then depends on two things: how concentrated supply is and how small the elasticity is. If short-run demand elasticity is small, a modest supply shortfall requires a large price jump. In October 1973 the OAPEC members declared an embargo. On 22 December 1973 the Gulf producers raised the posted price of the marker crude from 5.12 to 11.65 dollars a barrel, effective 1 January 1974. The mechanism repeated in European gas in 2022: the front-month TTF contract reached around 319 euros per MWh on 26 August 2022. And today the vulnerability has moved to the refining of critical materials: in 2024, according to the International Energy Agency, China refined about 91% of rare earths and produced 94% of sintered magnets. The limit: elasticity rises over time, because people and firms adapt, and stocks and substitution cushion the shock. The 1973 embargo worked less well as a ‘selective’ weapon than as a price shock, because oil is redirected among buyers. For the forecaster, a market with concentrated supply, small stocks and small elasticity produces violent prices from small shocks; when stocks are large, the shock fades.
Safety and certainty in oil lie in variety, and in variety alone.Winston Churchill, prim-lord al Amiralității, Camera Comunelor, 17 iulie 1913 (Hansard, vol. 55), dezbaterea despre trecerea flotei pe păcură; citat și de Daniel Yergin în «The Prize» (1991) (traducerea noastră)
Why it matters The TTF contract reached around 319 euros per MWh in the summer of 2022, a level nobody considered plausible a year earlier, precisely because gas demand elasticity was small.
In full →Collapse ↑
Worked example: 1973 and a relation you can calculate. On 22 December 1973 the Gulf producing states decided to raise the posted price of the marker crude from 5.12 to 11.65 dollars a barrel, effective 1 January 1974. The rise: (11.65 − 5.12) / 5.12 = 6.53 / 5.12 ≈ 1.275, that is +127.5%. The posted price is not the spot-market price, but it shows the order of magnitude: in about two months from the start of the embargo, the reference price more than doubled.
How large must the supply shortfall be for such a jump? With short-run demand elasticity ε (negative), the percentage price change needed to balance the market after a shortfall of ΔQ/Q is, to a first approximation, −(ΔQ/Q) / ε. Suppose, for the example, an effective shortfall of 7% (an illustrative figure, not the historical one). If ε = −0.1: 0.07 / 0.1 = 0.70, so +70%. If ε = −0.3: 0.07 / 0.3 ≈ 0.233, so +23%. If ε = −0.05: 0.07 / 0.05 = 1.40, so +140%.
A single parameter, unknown in real time, changes the answer about sixfold (from 23% to 140%). The linear approximation does not hold for large shocks, where the shape of the curve matters, but the order of magnitude is enough to see the logic. This is also why price forecasts during an energy shock are so badly calibrated: they depend on a parameter nobody observes until after the event.
In European gas in 2022 the mechanism was the same, with one difference: stocks. The front-month TTF contract reached around 319 euros per MWh on 26 August 2022; as storage filled and consumption fell, effective elasticity rose, and the price came back.
For the forecaster: the reference class is made up of energy supply shocks with a clear date (1973, 1979, 1990, 2022) and is counted by what fraction of the pre-shock price was reached at the peak. With only four or five cases, fine base rates cannot be made; wide intervals can. For critical materials there is not yet a class of episodes: Chinese export controls are too few and too recent.
Indicators to watch, with the public source. The level of gas stocks in Europe: the AGSI+ platform of Gas Infrastructure Europe, daily. Contract prices: ICE Endex (TTF) for gas, ICE for Brent. The oil supply and stocks situation: the monthly report of the International Energy Agency and the data of the US Energy Information Administration. For critical materials: the IEA's Global Critical Minerals Outlook and the USGS Mineral Commodity Summaries, both annual.
The typical trap: confusing the posted, administered price with the market price, or one contract month with another. A ‘record high’ figure without the contract specified cannot be resolved. The second trap: treating supply concentration as the only variable; without small stocks and small elasticity, concentration does not produce violent prices.
A scorable example question, illustrative: ‘Will the front-month TTF contract close above 100 €/MWh at ICE Endex in at least one session between 1 January and 31 December 2027?’ Resolution: the series of daily closes published by ICE Endex; the contract is identified by the ‘front month’ rule on the date of the session; the closing price counts, not the intraday high.
- 10
A negotiator plays simultaneously at the international table and at home: an agreement is possible only if it lies in the set that can be ratified on both sides, and whoever has hands tied at home may win more at the table.
Diplomacy and Domestic Politics: The Logic of Two-Level Games · Robert D. Putnam · 1988
In 1988, in Diplomacy and Domestic Politics: The Logic of Two-Level Games (International Organization), Robert Putnam proposed that we stop asking whether domestic politics determines foreign policy or the reverse, and instead picture a negotiator playing two games at once. At level I he negotiates with foreign counterparts; at level II he must obtain ratification at home, formal or informal. The set of agreements the domestic environment would accept is called the win-set. An agreement exists only if the win-sets of the two sides overlap, and the larger a win-set, the more likely the agreement. The mechanism has a counterintuitive consequence: a negotiator with a small win-set, that is, with a difficult ratification at home, can obtain bigger concessions at the table, because the partner knows he cannot ask for more anyway. The cost is the risk of failure. A historical example: on 19 March 1920 the US Senate voted 49 to 35 for the Treaty of Versailles, with Lodge's reservations, but 7 votes were missing for the two-thirds majority. Other works complete the framework: Bueno de Mesquita and colleagues, in The Logic of Political Survival (2003), show that a leader survives by satisfying the winning coalition within the selectorate, and James Fearon (1994) describes audience costs: a leader who has threatened publicly suffers at home if he backs down. Limits: the win-set is not observed, it is inferred. Audience costs have been challenged empirically, and regimes are not monolithic. For the forecaster, the question is who ratifies, by what majority and when; the parliamentary calendar and the arithmetic of majorities are public information.
Domestic politics and international relations are often somehow entangled, but our theories have not yet sorted out the puzzling tangle.Robert D. Putnam, «Diplomacy and Domestic Politics: The Logic of Two-Level Games», International Organization, vol. 42, nr. 3, 1988, pp. 427–460, propoziția de deschidere redată după descrieri secundare, neverificată pe PDF (traducerea noastră)
Why it matters Any international agreement that requires a parliament's ratification, from trade treaties to defence procurement, depends on domestic arithmetic as much as on the external negotiation.
In full →Collapse ↑
Worked example: the US Senate and the Treaty of Versailles, 1919–1920. The treaty required ratification by two thirds of the senators present. On 19 November 1919 two votes took place. The resolution to ratify with Henry Cabot Lodge's reservations was defeated 39 to 55. With 94 votes cast, two thirds means 94 × 2/3 = 62.67, so the threshold was 63 ‘yes’ votes: 24 were missing. The resolution to ratify without reservations, supported by President Wilson, was defeated 38 to 53. With 91 votes, the threshold was 60.67, so 61: 23 were missing.
Read through Putnam: the Senate's win-set included the treaty with reservations, and Wilson's win-set included the treaty without them, but not with them. The two sets had no common point wide enough. On 19 March 1920, in a second vote, the resolution with Lodge's reservations received 49 ‘yes’ votes and 35 ‘no’ votes. Out of 84 votes, two thirds means 56, so 7 votes were missing for ratification.
The observation from the exercise: all three votes failed on arithmetic, not on diplomacy. A forecaster who had counted the votes in 1919 on both variants would have seen that neither reaches the two-thirds threshold, whatever the two camps said. The lesson is not that ‘the opposition won’ but that the ratification rule (two thirds) made the win-set very small, and when the domestic win-set is small, a successful negotiation at level I guarantees nothing at level II. The check is simple: count the votes required, then count the votes available.
For the forecaster: the reference class is the ratification of treaties in systems with qualified majorities. It is built by counting, for each treaty submitted in a given interval, how many were approved; for the US, treaty documents and votes are public on Congress.gov and on the Senate's site. For Romania, on the websites of the Chamber of Deputies and the Senate. If you do not have such a count, do not invent a rate: say that you are using the arithmetic of majorities, not a base rate.
Indicators to watch, with the public source. The calendar and agenda of the ratifying body: the website of the parliament concerned. The arithmetic of majorities: the number of seats by group, from the parliament itself. Opinion polls in the country concerned: Eurobarometer or national polling institutions. Leaders' statements on red lines: official communiqués, read as a possible strategy, not as fact.
The typical trap: taking ‘tied hands’ as real. The statement ‘parliament would not vote for it’ may be true or may be a negotiating move; you check with the votes, not with the rhetoric. The second trap: forgetting that a signed agreement is not a ratified one. A signature at level I says nothing about level II.
A scorable example question, illustrative: ‘Will the US Senate approve, by a final vote with at least 67 ‘yes’ votes, the resolution of ratification of a treaty submitted to the Senate on 1 March 2027, by 31 December 2027, according to the roll-call record published by the Senate?’ Resolution: the official roll-call record; if no vote takes place, the outcome is ‘no’; amendments do not change the criterion.
- 11
Organski and Allison tie war to a rising power approaching parity: 12 of 16 cases chosen by Allison ended in war, but a list selected by the author is not a base rate.
Destined for War: Can America and China Escape Thucydides's Trap? · Graham Allison · 2017
The idea that war threatens when a rising power catches up with a dominant one comes from A.F.K. Organski, who in World Politics (1958) formulated power transition theory: the international order is a hierarchy, and the danger is greatest when a power dissatisfied with the existing rules approaches parity with the dominant one. Graham Allison popularised the idea under the name ‘Thucydides's Trap’ and made it the central thesis of Destined for War (2017): of 16 cases in the past 500 years in which a rising power threatened to displace a ruling one, 12 ended in war. The mechanism invoked: the rising power demands status and influence, the dominant one fears loss, alliances tie both sides' hands, and a minor crisis can trigger a major conflict. The most discussed case is the Anglo-German rivalry before the First World War. The cases without war include the rivalry between Portugal and Spain at the end of the fifteenth century and the Cold War between the US and the USSR. The criticism is strong. Joseph Nye and others argue that some cases are misidentified, and the list was compiled by the author, after the theme had been chosen. With 16 cases and a sample largely European, any rate is very uncertain. For the forecaster, the lesson is methodological: a reference class of 16 cases chosen by the author is not a base rate. The useful benchmark is an explicit inclusion rule, applied before seeing the outcomes.
Intentions aside, when a rising power threatens to displace a ruling power, the resulting structural stress makes a violent clash the rule, not the exception.Graham Allison, «Destined for War: Can America and China Escape Thucydides's Trap?», Houghton Mifflin Harcourt, 2017, formulare redată după citări secundare, pagina neverificată (traducerea noastră)
Why it matters The US–China rivalry is often framed as ‘Thucydides's Trap’, but a serious forecast begins by asking which reference class, built by which rule, stands behind the figure of 12 out of 16.
In full →Collapse ↑
Worked example: what the 12 out of 16 say, and what they do not. Take Allison's figure: 16 cases in which a rising power threatened to displace a ruling one, of which 12 ended in war. The fraction: 12 / 16 = 0.75.
First step: if the probability of war were 50% in each case, what are the chances of seeing at least 12 wars out of 16? Count the combinations: C(16,12) = 1,820, C(16,13) = 560, C(16,14) = 120, C(16,15) = 16, C(16,16) = 1, 2,517 in all. Divided by 2^16 = 65,536, that gives 0.0384, or 3.8%. Provisional conclusion: if the list were a random sample, 12 out of 16 would contradict a 50% rate at a conventional threshold.
Second step: the interval. The 95% Wilson interval for 12 out of 16 covers roughly 51%–90%. With 16 cases, any rate has a large uncertainty: from ‘a little over half’ to ‘almost nine in ten’.
Third step, the decisive one: selection. The list was compiled by the author, after choosing the theme. Suppose, as an exercise, that a wider list added 8 cases without war, for example parity rivalries that were resolved peacefully and that the author did not identify. The rate would be 12 / 24 = 0.50. We do not say that this is what happened: we say that a single inclusion decision moves the rate by 25 percentage points. Joseph Nye disputes precisely the identification of some cases.
What remains: a research hypothesis, not a base rate. To become a rate, you need a rule written in advance, for example: ‘pairs of states in which the rising power has at least 80% of the dominant power's capability, measured by a public index’. With the rule applied across the whole period, the list is rebuilt, and the rate is reported with an interval.
For the forecaster: the reference class is built from the inclusion rule, not from the catalogue. A serious source for capabilities is the Correlates of War project, with the CINC index (Composite Index of National Capability); for conflicts, the database of militarised interstate disputes. The rate is counted over pairs of states and years, not over ‘rivalries’ chosen by narrative. If the rule yields too few cases, say so: a wide interval, not a smooth figure.
Indicators to watch, with the public source. The ratio of the challenger's GDP to that of the dominant power: the IMF, World Economic Outlook, twice a year. Military spending: SIPRI, annual. Military incidents between the two sides: defence ministry statements and independent monitoring sources, read with care. Alliances and commitments: the public texts of treaties.
The typical trap: treating a historical analogy as law. ‘1914’ is one case among many, and a similar path does not guarantee a similar outcome. The second trap: treating the absence of war so far as proof that the thesis is false; a rate of 75% leaves 25% for peace, and a small sample refutes nothing.
A scorable example question, illustrative: ‘Will SIPRI report, in its database published in April 2027, that China's military spending for 2026, in constant dollars, exceeds 40% of that of the US?’ Resolution: the constant-dollar series in the SIPRI database; the ratio is computed from the values published in the April 2027 edition; a later revision of the series does not change the outcome.
- 12
Population moves slowly and predictably: those who will be 40 in twenty years are already born, so demography is the one social variable with solid long-term forecasts, apart from migration and future fertility.
World Population Prospects 2024 · Organizația Națiunilor Unite, Departamentul pentru Afaceri Economice și Sociale (Divizia Populație) · 2024
The demographic transition, formulated from the observations of Warren Thompson (1929) and developed by Frank Notestein (1945), describes the shift from high mortality and high birth rates to low mortality and low birth rates. Mortality falls first, population grows, then birth rates fall and growth slows. The key benchmark is the replacement fertility rate, about 2.1 children per woman in low-mortality countries: below it, each generation is smaller than the one before. The dependency ratio, the number of people over 65 relative to those aged 15 to 64, shows the pressure on pensions and health systems. The mechanism is cohort inertia. A person born in 2026 will be 20 in 2046; those who will be 50 in 2046 are already alive today, and their forecast depends only on mortality and migration. Romania offers an example: Decree 770 of 1966, which banned abortion, almost doubled the number of births in 1967, and the large cohorts of that time are now moving toward retirement age. Between the censuses of 1992 and 2021 the population fell from 22.81 to 19.05 million. Limits: migration and future fertility. In World Population Prospects 2024, where the world's population is projected to peak at about 10.3 billion in the 2080s, the projected population for 2100 is 6% lower than in the edition of ten years earlier, mainly because fertility fell faster than expected, especially in China. For the forecaster, the age structure at 20 years is very safe for those over 20 and very uncertain for those not yet born.
The world's population is projected to grow from 8.2 billion in 2024 to a peak of around 10.3 billion in the mid-2080s, then decline slightly, to around 10.2 billion by the end of the century.Organizația Națiunilor Unite, DESA, «World Population Prospects 2024», comunicat de presă din 11 iulie 2024, redat după relatări, nu cuvânt cu cuvânt (traducerea noastră)
Why it matters Between the 1992 and 2021 censuses Romania's population fell by about 3.76 million people, or 16.5%, and those who will be 50 in 2046 are already alive.
In full →Collapse ↑
Worked example: Romania between two censuses. The census of 7 January 1992 counted 22,810,035 people (the stable population). The 2021 census counted 19,053,815 people (the resident population), according to the National Institute of Statistics. The difference: 22,810,035 − 19,053,815 = 3,756,220. Relative to 1992: 3,756,220 / 22,810,035 ≈ 0.165, a fall of 16.5%. Over 29–30 years, that means an average fall of about 0.6% a year: ln(19,053,815 / 22,810,035) ≈ −0.180, and −0.180 / 29.9 ≈ −0.006.
Watch the definitions: in 1992 the stable population was counted, in 2021 the resident one. The two are not identical, so the 16.5% figure is an order of magnitude, not a precise measurement. A serious forecaster notes this in the criterion of the question and names the exact source.
The second calculation, replacement. If a country had a fertility of 1.7 children per woman (an illustrative value) and low mortality, the ratio 1.7 / 2.1 ≈ 0.81 shows that each generation is about 81% of the size of the preceding one, that is, about 19% smaller. Over a generation of roughly 30 years, the fall is visible but slow, and migration can offset or aggravate the result. The calculation is not a forecast but an order-of-magnitude test: if a scenario requires a faster fall than this, it must explain by what, for example emigration.
The third step: the cohort. Whoever is between 50 and 54 in 2046 was born between 1992 and 1996. The cohort exists; it can grow only through immigration and shrinks through mortality and emigration. That is why its 20-year forecast is safe, unlike that of children born in 2045, who do not yet exist. Any figure about them is a fertility scenario, not an observation.
For the forecaster: the reference class for demography is not a base rate of events but the historical errors of projections. It is built by comparing projections published 10–20 years ago with today's results: errors on total population are small over 10 years and grow with the horizon, and the largest come from migration and fertility. The United Nations' world projection for 2100 was revised down by 6% in the 2024 edition compared with that of a decade earlier: an order of magnitude for revisions.
Indicators to watch, with the public source. Annual births and deaths, the fertility rate: the National Institute of Statistics (TEMPO) and Eurostat. Migration flows: Eurostat and the immigration statistics of destination countries. Official projections: the United Nations, World Population Prospects, and Eurostat's projections. The old-age dependency ratio: Eurostat, by country.
The typical trap: treating a projection as a fact, without reading the fertility and migration assumptions it rests on. The second trap: comparing censuses with different definitions (stable and resident population) as though they counted the same thing.
A scorable example question, illustrative: ‘Will the number of live births in Romania in 2026, from the INS's first press release on 2026, be below 140,000?’ Resolution: the INS release published in 2027 on the natural movement of the population in 2026; the figure from the first publication is used, not later revisions; if the INS does not publish by 31 December 2027, the question is voided.
Chapter 3The mechanics of the macroeconomy
- 13
GDP is an accounting identity, not a theory: what is produced is bought by someone, and a current-account deficit is, just as exactly, the gap between what a country invests and what it saves.
National Income, 1929–1932 · Simon Kuznets · 1934
National accounts are an invention of the 1930s crisis. In 1934 Simon Kuznets handed the US Senate the first coherent estimate of national income for 1929–1932, and in 1940 John Maynard Keynes used estimates of the same kind in ‘How to Pay for the War’ to show how much of output could go to the war effort without inflation. From these accounts comes an identity: gross domestic product Y equals consumption C plus investment I plus government spending G plus net exports NX. It is not a hypothesis to test but a definition: everything that is produced is bought by someone. Subtract net taxes T and consumption from Y and what remains is private saving S, and rearranging gives NX = (S − I) + (T − G): the current account (approximated here by NX) is the sum of the private sector’s saving surplus and the budget balance. A country with a negative current account spends and invests more than it produces, and the gap is financed from abroad; the financial account in the balance of payments shows the same amount with the opposite sign, up to errors and omissions. This is where the phrase ‘twin deficits’ comes from, coined in the United States in the 1980s, when the federal deficit and the external deficit grew together. The identity does not say what causes what, however. In 2007 Spain ran a budget surplus and yet a current-account deficit of about 10% of GDP, because the deficit belonged to the private sector, financed by foreign credit. Kuznets himself warned that national income does not measure welfare. For a forecaster, identities are a coherence test: any story about growth or about an exchange rate must show who spends, who saves and who finances the difference.
The welfare of a nation can, therefore, scarcely be inferred from a measurement of national income as defined above.Simon Kuznets, «National Income, 1929–1932», raport către Senatul SUA (Senate Document No. 124, 73rd Congress, 2nd Session), 1934, p. 7 (redarea română: traducerea noastră)
Why it matters At the end of 2024 Romania had the largest government deficit in the EU (9.3% of GDP) and a current-account deficit of about 8.4% of GDP; the identity shows the two figures cannot be read separately.
In full →Collapse ↑
Worked example: an open economy with round numbers (fictional currency units). Consumption C = 700, investment I = 200, government spending G = 250, exports 150 and imports 200, so NX = −50. Net taxes T = 220.
Y = C + I + G + NX = 700 + 200 + 250 + (−50) = 1,100 S = Y − T − C = 1,100 − 220 − 700 = 180 S − I = 180 − 200 = −20 T − G = 220 − 250 = −30 (S − I) + (T − G) = −20 + (−30) = −50 = NXThe external deficit of 50 is not a separate fact: it is the sum of two shortfalls, 20 in the private sector (it invests more than it saves) and 30 in the state. The financing comes through the financial account: non-residents buy a net 50 of domestic assets (loans, securities, equity), the same amount with the opposite sign. In real data small differences appear, from errors and omissions and from capital transfers.
First shock. The state raises spending by 30 (G = 280), with S and I unchanged. Then T − G = −60, S − I stays at −20, and the current account becomes −80: the external deficit grows by exactly 30, and Y = 700 + 200 + 280 − 80 = 1,100. Second shock: the same rise in G, but households save 30 more (C = 670, so S = 210). Now S − I = +10, T − G = −60, and the current account stays at −50. The same fiscal measure, two different external outcomes: what decides is the private sector’s reaction.
The last step matters for forecasting. If foreign investors stop financing the 50, the current account must reach zero. With Y staying at 1,100, C + I + G would have to fall from 1,150 to 1,100. That is, seen from the inside, the ‘sudden stop’ we meet in lesson 20: someone, somewhere in the economy, spends 50 less.
For the forecaster: the reference class. There is no single base rate for ‘a large external deficit closes abruptly’, but it can be counted: take from the World Development Indicators (World Bank) or the World Economic Outlook (IMF) every country–year pair with a current account below −5% of GDP for three years running, and count how many saw, within the next two years, an improvement of at least 4 percentage points. The thresholds are fixed before counting, not after choosing the countries.
Romania’s position at the end of 2024: a general-government ESA deficit of 9.3% of GDP (Eurostat, the largest in the EU) and a current-account deficit of about 8.4% of GDP (National Bank of Romania, BNR). The identity shows the private sector was close to balance, so the external deficit was largely the public deficit; the relation is only approximate, because EU capital transfers and errors bend it.
Indicators to watch: the monthly balance of payments and current-account balance (BNR); the ESA deficit, notified to Eurostat in April and October (the National Institute of Statistics, INS, and the Ministry of Finance); foreign direct investment and portfolio flows, as the measure of financing (BNR); goods imports (INS).
The trap: reading the identity as causation. A larger budget deficit does not automatically produce a larger external deficit; it depends on what private saving and investment do. The second trap: revisions, because BNR and INS restate whole series, and the score must be tied to the first publication.
A scorable question (illustrative): ‘Will the BNR report, in its press release on the 2026 balance of payments published by 31 March 2027, a current-account deficit larger than 7.0% of GDP?’ Resolution source: the first value published in the BNR release; later revisions do not count.
- 14
In the long run prices follow money relative to output, but the link runs through velocity, expectations and supply shocks, so it is not an automatism you can read off a single series.
The Counter-Revolution in Monetary Theory · Milton Friedman · 1970
The equation of exchange, MV = PY, was formulated by Irving Fisher in 1911: the quantity of money M times the velocity of circulation V equals the price level P times real output Y. As an identity it is true regardless; it becomes a theory when V is assumed stable and Y is set by real factors, in which case prices follow money. Milton Friedman summed up the position in 1970, in his Wincott lecture ‘The Counter-Revolution in Monetary Theory’: inflation is a monetary phenomenon. The rival was the Phillips curve: in 1958 A. W. Phillips found, in British data for 1861–1957, an inverse link between unemployment and wage growth. Edmund Phelps (1967) and Friedman (1968) showed that the link vanishes in the long run if people adjust their expectations, and the 1970s proved them right: high inflation and high unemployment at once. Paul Volcker tightened monetary policy from October 1979; American inflation fell from 14.8% in March 1980 to an average of 3.2% in 1983, at the price of unemployment of 10.8% in December 1982. The 2021–2023 wave took another form: supply shocks in energy and logistics, and the euro-area annual rate reached 10.6% in October 2022. The limits are real. Velocity V did not stay stable after 1980, and central banks gradually abandoned money-supply targets. Supply shocks can raise prices without money growing faster. For a forecaster two things matter: the anchoring of expectations (when they come loose, disinflation costs more) and the lag, of the order of 12–24 months, between policy and prices.
Inflation is always and everywhere a monetary phenomenon in the sense that it is and can be produced only by a more rapid increase in the quantity of money than in output.Milton Friedman, «The Counter-Revolution in Monetary Theory» (prelegerea Wincott, Londra, 16 septembrie 1970), Institute of Economic Affairs, Occasional Paper 33, 1970 (redarea română: traducerea noastră)
Why it matters The euro-area annual inflation rate has topped 10% in only two months since the currency was launched, October and November 2022, and central banks answered with the fastest monetary tightening in decades.
In full →Collapse ↑
Worked example: the equation of exchange with numbers, then two real episodes. Let M = 1,000 and V = 2, so PY = 2,000; assume Y = 2,000 and P = 1.00. Money grows by 10%, velocity stays at 2, and real output grows by 3%.
new M = 1,000 × 1.10 = 1,100 new PY = 1,100 × 2 = 2,200 new Y = 2,000 × 1.03 = 2,060 new P = 2,200 / 2,060 = 1.0680 → prices +6.8% (quick approximation: 10 − 3 = 7 points) If V falls by 2% (V = 1.96): PY = 1,100 × 1.96 = 2,156 new P = 2,156 / 2,060 = 1.0466 → prices +4.7%The same stock of money gives 6.8% or 4.7% depending on whether velocity is stable. That is why simple monetarism has held as an order of magnitude over decades, not as a forecast over quarters.
First episode: Volcker. American inflation (CPI, 12-month) had reached 14.8% in March 1980. The Federal Reserve moved in October 1979 to targets for bank reserves, and the monthly average of the federal funds rate climbed to 19.1% in June 1981. The cost was a double recession: unemployment reached 10.8% in December 1982, but average inflation in 1983 was 3.2%. Second episode: the euro area. The annual rate of the harmonised index (HICP) rose to 10.6% in October 2022, dominated by energy, and the ECB raised its deposit rate from −0.50% in July 2022 to 4.00% in September 2023. The first case was largely an announced monetary shock; the second, a supply shock met with monetary tightening once expectations had begun to move. The same equation, then, with other variables in motion: M in the first case, supply Y and expectations in the second, which is exactly what a single money series does not show.
For the forecaster: the base rate. Since the launch of the euro, the euro-area annual HICP rate has exceeded 10% in two months: October 2022 (10.6%) and November 2022 (10.1%); a jump of that kind is a tail event, not an ordinary move. For lower thresholds the reference class is built from the monthly Eurostat series, counting the months in which the annual rate passed the chosen threshold, but only from periods with a comparable monetary regime.
Indicators to watch: the consumer price index and the core component, excluding energy and food (Eurostat for the euro area, INS for Romania); negotiated wage growth and unit labour costs (ECB, Eurostat); inflation expectations, from the ECB’s Consumer Expectations Survey or the University of Michigan survey; the monetary aggregate M3 (ECB), bearing in mind that its relation to prices is weak in the short run.
The trap: base effects. A tax or a price cap that enters the index shows up in the annual rate for 12 months and then drops out abruptly, so the rate falls with no real change of trend. A second trap: forecasting annual inflation with a ‘M is growing, so prices will grow’ rule without asking how much V has moved.
A scorable question (illustrative): ‘Will Romania’s annual CPI inflation rate, published by INS for December 2026, be below 4.0%?’ Resolution source: the INS consumer-price release for December 2026, the 12-month rate, first published value; later revisions do not count.
- 15
The 1993 Taylor rule ties the policy rate to inflation and the output gap; it is not the law by which central banks decide but a benchmark for measuring how tight or loose a policy is.
The Role of Monetary Policy · Milton Friedman · 1968
In 1993 John Taylor published ‘Discretion versus Policy Rules in Practice’ in the Carnegie-Rochester series, with a simple rule that described the Federal Reserve’s interest rate for 1987–1992 fairly well: i = π + 0.5·y + 0.5·(π − 2) + 2. Here i is the nominal policy rate, π is inflation over the previous four quarters, y is the output gap in percent, and the two at the end are the inflation target and the neutral real rate. It is not the law by which central banks decide but a benchmark. The mechanism lies in the inflation coefficient: π enters once directly and 0.5 times in the gap from the target, so for every extra point of inflation the rate rises by 1.5 points and the real rate goes up (the Taylor principle). When the rule is violated, the real rate falls as inflation rises, as in the 1970s according to some estimates. In July 2022, with euro-area inflation at 8.6% in June, the ECB raised its deposit rate from −0.50% to 0.00%; it reached 4.00% only in September 2023. The effect is transmitted through credit, asset prices, the exchange rate and expectations, with Friedman’s ‘long and variable’ lags. The main limit is that the neutral rate cannot be observed: it is estimated, and the estimates fell after 2008 and were revised after 2022. The output gap is revised too, sometimes by whole points. For a forecaster, the rule is a plausibility band: if the central bank stays far below it, the useful question is what the bank knows or fears, not whether the rule is wrong.
The first and most important lesson that history teaches about what monetary policy can do—and it is a lesson of the most profound importance—is that monetary policy can prevent money itself from being a major source of economic disturbance.Milton Friedman, «The Role of Monetary Policy», American Economic Review, vol. 58, nr. 1, martie 1968 (discursul prezidențial AEA din decembrie 1967), p. 12 (redarea română: traducerea noastră)
Why it matters In mid-2022 the Taylor rule, fed with the inflation of the day, pointed to rates of the order of 14% while the ECB’s rate was 0%; the gap foreshadowed the steepest hiking cycle in the euro’s history.
In full →Collapse ↑
Worked example: the Taylor rule with round numbers, then a real case. The formula: i = π + 0.5·y + 0.5·(π − 2) + 2, with π inflation, y the output gap in percent, the target 2% and the neutral real rate 2%.
Case π y calculation i A on target 2 0 2 + 0 + 0 + 2 4.0% B inflation 4, gap -1 4 -1 4 − 0.5 + 1.0 + 2 6.5% C inflation 6, gap -1 6 -1 6 − 0.5 + 2.0 + 2 9.5% D as B, neutral rate 1 4 -1 4 − 0.5 + 1.0 + 1 5.5%From B to C inflation rises by 2 points and the indicated rate by 3 (1.5 × 2). The real rate (i − π) goes from 6.5 − 4 = 2.5 to 9.5 − 6 = 3.5: it rises by one point. Case D shows how much the result hangs on an unobservable number: if the neutral rate is 1% and not 2%, the same economy calls for 1 point less.
The real case. In June 2022 the euro-area annual inflation rate was 8.6%. With a zero output gap and the same formula: i = 8.6 + 0 + 0.5 × (8.6 − 2) + 2 = 8.6 + 3.3 + 2 = 13.9%. On 21 July 2022 the ECB decided on a deposit rate of 0.00%, almost 14 points below the benchmark. The rule overstated, of course: inflation at the time was largely energy, and part of it was temporary; on core inflation, much lower, the calculation gives a much smaller result. Still, when policy stands this far from the rule, the direction of the next meetings can be forecast: the ECB went from 0.00% to 4.00% in 14 months. The rule did not predict the level, but the direction and the pressure.
For the forecaster: base rate and signals. The base rate for a central bank’s decisions is built by counting meetings: for the last 100 monetary-policy meetings of the chosen bank, the share with a rate change against those with a hold. Most meetings are holds as a rule, and central banks prepare the market before a move; that is why the best conditional base rate is the rate path implied by market prices, not the historical average.
Indicators to watch: HICP and core inflation (Eurostat); the quarterly macroeconomic projections of ECB staff and the statements after meetings (ECB); the curve of rates implied by €STR contracts, that is, what the market expects (published by the ECB and by market data providers); output-gap estimates (IMF, European Commission, CBO for the United States).
The trap: the Taylor rule gives a range, not a number. The choice of inflation (headline or core), of the neutral rate and of the gap can move the result by 3–5 points; whoever uses it as a point forecast picks the result after seeing the decision. A second trap: confusing ‘should’ with ‘will’: a central bank reacts to its own projection, not to the formula.
A scorable question (illustrative): ‘Will the ECB deposit facility rate, after the last monetary-policy meeting of the first half of 2027, be below 2.00%?’ Resolution source: the ECB page on key interest rates and the statement of that meeting; if the ECB changes its key instrument, use the rate at which it remunerates overnight deposits.
- 16
The spread between the 10-year and 3-month yields is the best-studied indicator of US recessions, but eight hits remain a small sample, and the 2022–2024 inversion showed that the signal is a probability, not a date.
Science and Stocks: Sense and Nonsense · Paul A. Samuelson · 1966
In 1996 Arturo Estrella and Frederic Mishkin showed, in an issue of the New York Fed’s ‘Current Issues in Economics and Finance’, that the spread between the 10-year and the 3-month US Treasury yields predicts recessions better than other financial or macroeconomic indicators, at horizons of two to six quarters. The spread goes into a probit model, which turns a number into a probability of recession four quarters ahead. The mechanism: the 10-year yield is the average of expected short rates plus a term premium; when the market thinks the central bank will cut rates because the economy is weakening, the spread narrows or turns negative. At the same time, high short rates squeeze the margin of banks, which borrow short and lend long, so credit tightens. The inversions of 2000, 2006–2007 and 2019 were followed by recessions that began in March 2001, December 2007 and February 2020. Recessions are dated by the NBER, with a delay: the December 2007 peak was announced in December 2008, the February 2020 one in June 2020. The limits: eight NBER recessions since 1968 are a small sample, and the indicator has false alarms (1966 is the one most often cited; some authors add 1998). The 2022–2024 inversion, lasting about two years, was not followed, by the end of 2024, by any NBER-dated recession; central-bank asset purchases can distort the term premium. For a forecaster: the spread gives a probability with a wide lag, not a date.
Wall Street indexes predicted nine out of the last five recessions! And its mistakes were beauties.Paul A. Samuelson, «Science and Stocks», rubrica din Newsweek, 19 septembrie 1966, p. 92 (redarea română: traducerea noastră)
Why it matters The 10-year–3-month spread stayed negative for about two years, from autumn 2022 into 2024, the longest inversion in decades, with no NBER recession by the end of 2024: an indicator with eight hits was no longer a verdict.
In full →Collapse ↑
Worked example: how to read a spread, and how strong an indicator with eight hits really is. Estrella and Mishkin (1996, data 1960–1995) publish a table linking the 10-year–3-month spread to the probability of a recession four quarters ahead.
10-year – 3-month spread (percentage points) Probability of recession 4 quarters ahead +1.21 5% −0.82 50% −2.18 (average for Q1 1981) 86.5%Reading it: at a spread of +1.21 the model gives 5%; at −0.82, 50%; and the −2.18 average of the first quarter of 1981 announced 86.5% for the first quarter of 1982, which was indeed in recession (July 1981 – November 1982). The spread is computed every day from two public yields: the example works at any date.
Now the sample. NBER recessions since 1968 number eight: 1969–1970, 1973–1975, 1980, 1981–1982, 1990–1991, 2001, 2007–2009 and 2020. Suppose, generously, that the indicator caught all of them. If the true hit rate were p, the probability of seeing eight out of eight is p to the power 8. How low can p fall before eight out of eight becomes improbable (below 5%)? From p^8 = 0.05 we get p = 0.05^(1/8): the natural logarithm of 0.05 is −2.996, divided by 8 gives −0.3745, and its exponential is 0.688. Conclusion: eight hits out of eight are compatible with a true rate of about 69%, and the false alarms (1966, 1998 according to some authors, 2022–2024 on data to the end of 2024) are not in this calculation. A small sample does not become large because it got lucky eight times.
For the forecaster: base rate and indicators. The reference class is the list of NBER-dated recessions: between 1948 and 2020 there are 12 recession starts (1948, 1953, 1957, 1960, 1969, 1973, 1980, 1981, 1990, 2001, 2007, 2020), about one every six years on average. Over a 12-month window the crude base rate is about 12 / 73 ≈ 16%; it is a starting point that the spread moves up or down, not a current forecast.
Indicators to watch: the 10-year–3-month spread, computed from US Treasury yields published by the Federal Reserve (the H.15 series, redistributed by FRED); the recession probability published monthly by the New York Fed on the basis of the spread; initial claims for unemployment insurance (Department of Labor); announcements of the NBER Business Cycle Dating Committee.
The trap: confusing inversion with the start of the recession. The interval between the first inversion and the cycle peak has ranged from a few months to almost two years, and inversions have also been followed by periods with no recession. A second trap: NBER dating comes late, sometimes half a year after the peak, so a question tied to it must have a wide resolution window.
A scorable question (illustrative): ‘Will the daily 10-year–3-month spread (the T10Y3M series on FRED) be negative on at least one business day between 1 January and 31 December 2027?’ Resolution source: the T10Y3M series on FRED, values available on 15 January 2028; days with no value are not counted.
- 17
The path of public debt depends on the gap between the interest rate and nominal growth and on the primary balance; thresholds such as 60% or 90% of GDP are conventions, not laws, and the arithmetic says more than they do.
Public Debt and Low Interest Rates · Olivier Blanchard · 2019
The dynamics of public debt are arithmetic. With d debt as a percentage of GDP, r the average nominal interest rate, g nominal GDP growth and sp the primary balance (the surplus before interest), the annual change is Δd = (r − g)/(1 + g) · d − sp. Debt rises if the interest rate exceeds growth or if the primary deficit is large. Olivier Blanchard, in his 2019 AEA presidential address ‘Public Debt and Low Interest Rates’, showed that in the United States the safe rate had stayed below growth for a long time, which makes debt easier to carry than is commonly believed. The historical example is the postwar one: US federal debt held by the public fell from about 106% of GDP in 1946 to about 23% in 1974, not through huge surpluses but through high nominal growth and low real interest rates. In 2010 Carmen Reinhart and Kenneth Rogoff argued in ‘Growth in a Time of Debt’ that above 90% of GDP growth falls sharply; in 2013 Thomas Herndon, Michael Ash and Robert Pollin found a spreadsheet error, exclusions of years and weighting choices, and average growth at debt above 90% came out at +2.2%, not −0.1%. The Maastricht Treaty sets 3% for the deficit and 60% for debt, two figures linked arithmetically: at 5% nominal growth, a 3% deficit stabilises debt at 3/0.05 = 60%. Limits: r and g are not constant, and r depends on market confidence: when investors demand a risk premium, the rate rises just when debt is high. For a forecaster the useful indicator is not a magic threshold but the gap r − g and the primary balance, together with the debt structure (maturities, currency).
Put bluntly, public debt may have no fiscal cost.Olivier Blanchard, «Public Debt and Low Interest Rates», American Economic Review, vol. 109, nr. 4, 2019, pp. 1197–1229, rezumat (versiunea de lucru NBER nr. 25621) (redarea română: traducerea noastră)
Why it matters At the end of 2024 Romania’s government debt was about 55% of GDP, below the 60% threshold, but with an ESA deficit of 9.3% of GDP: the arithmetic of the trajectory matters more than today’s level.
In full →Collapse ↑
Worked example: the path of debt over five years, exact calculation. Start with d = 60% of GDP, an average nominal interest rate r = 6%, nominal growth g = 7% and a primary deficit of 4% of GDP (sp = −4). The formula: next d = d · (1 + r)/(1 + g) − sp. The factor 1.06 / 1.07 = 0.990654.
year calculation debt (% of GDP) 0 - 60.0 1 60.0 × 0.990654 + 4 63.4 2 63.44 × 0.990654 + 4 66.8 3 66.85 × 0.990654 + 4 70.2 4 70.22 × 0.990654 + 4 73.6 5 73.57 × 0.990654 + 4 76.9(The calculation was done with full decimals and rounded only at the end: 63.439; 66.846; 70.222; 73.565; 76.878.) Debt climbs by about 3.4 points a year even though the interest rate is below growth: the primary deficit of 4 points far outweighs the gain of 0.56 points a year from the gap r − g, that is, (0.06 − 0.07) / 1.07 × 60 ≈ −0.56.
What primary balance would have stabilised debt at 60%? The one for which Δd = 0: sp = (r − g)/(1 + g) · d = −0.56, a primary deficit of 0.56% of GDP. With a deficit of 4%, the country is 3.4 points a year away from stabilisation. Had the interest rate been 8% instead of 6% (g staying at 7%), the factor becomes 1.08 / 1.07 = 1.009346, and after five years debt reaches about 83.2%, that is 6.4 points higher. Two points more interest, a little over 6 points more debt in five years: that is why the path is forecast from r − g, not from today’s level. The rule to remember: any primary deficit larger than (r − g)/(1 + g) · d pushes debt up, however small the interest rate may look.
For the forecaster: base rate and indicators. At the end of 2024 Romania’s government debt was about 55% of GDP (Eurostat), below the 60% reference, while the ESA deficit was 9.3% of GDP, the largest in the EU. The Ministry of Finance also reported a ‘cash’ deficit of 8.65%: the difference between the cash basis and the accrual basis (ESA) is a classic resolution trap.
The reference class is counted from Eurostat’s excessive-deficit notifications: how many member states had, in a given year, a deficit above 3% of GDP and how many of them brought it below the threshold within the next two years; for debt thresholds the IMF’s Historical Public Debt Database can be used. No probability is attached before counting.
Indicators to watch: Eurostat’s government deficit and debt notifications (April and October); monthly budget execution (Ministry of Finance); the 10-year government bond yield and the average cost of debt (BNR, Ministry of Finance); nominal GDP growth (INS).
The trap: using a threshold as a law. 90% or 60% do not produce a crisis by themselves; what matters is who holds the debt, in which currency it is issued and how fast it rolls over. A second trap: comparing figures on different bases (cash, ESA, gross or net debt).
A scorable question (illustrative): ‘Will Eurostat report, in the April 2027 excessive-deficit notification, a Romanian general-government ESA deficit for 2026 above 6.0% of GDP?’ Resolution source: the Eurostat release of the April 2027 notification, the value for 2026, first published.
- 18
A fixed rate, free capital and an independent monetary policy cannot all exist at once; and the floating rate itself, over short horizons, is hard to forecast better than ‘today’s rate’.
Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample? · Richard A. Meese și Kenneth Rogoff · 1983
The monetary trilemma follows from the model formulated independently by Marcus Fleming (1962, IMF Staff Papers) and Robert Mundell (1963): a country cannot have at once a fixed exchange rate, free movement of capital and an independent monetary policy; it picks two of the three. With free capital and a fixed rate, the domestic interest rate is tied to that of the anchor country. The mechanism: if the market believes the fixed rate will not hold, defending it takes high interest rates and reserves. Britain joined the European Exchange Rate Mechanism in October 1990, at 2.95 marks to the pound, in a band of ±6%. With German rates high after unification, on 16 September 1992 the Bank of England raised its rate from 10% to 12% and then announced 15% the same day, but in the evening the government pulled sterling out of the mechanism; in the following weeks it adopted an inflation target. Rüdiger Dornbusch (1976, ‘Expectations and Exchange Rate Dynamics’) showed that, with sticky prices, the exchange rate can overshoot its new equilibrium before coming back. Uncovered interest parity says that the currency with the higher interest rate is expected to depreciate by the difference. In practice parity does not hold well: high-interest currencies usually depreciate less than the formula says, which is why carry trades exist. Meese and Rogoff (1983) showed that at horizons of 1–12 months no structural model beats a random walk. For a forecaster: exchange rates are not predicted point by point; regimes (anchor, band, defence) and breaking points are forecast.
The random walk model performs as well as any estimated model at one to twelve month horizons for the dollar/pound, dollar/mark, dollar/yen and trade-weighted dollar exchange rates.Richard A. Meese și Kenneth Rogoff, «Empirical Exchange Rate Models of the Seventies: Do They Fit Out of Sample?», Journal of International Economics, vol. 14, 1983, rezumat (redarea română: traducerea noastră)
Why it matters The BNR intervenes in the foreign-exchange market to limit the leu’s volatility: for a question about the exchange rate, the regime and the reserves say more than any structural model.
In full →Collapse ↑
Worked example: 16 September 1992, then the arithmetic of parity. Britain joined the European Exchange Rate Mechanism in October 1990 at 2.95 German marks to the pound, in a band of ±6%. The lower edge of the band, about 2.78 marks, lay at 2.78 / 2.95 − 1 ≈ −5.8% from the central parity. The trilemma reads directly: sterling had a fixed rate (the mechanism), free capital (exchange controls had been abolished in 1979) and wanted its own monetary policy, because the British economy was in recession in 1990–1991 and would have needed lower rates, while the Bundesbank held rates high after German unification.
On 16 September 1992 the Bank of England raised its rate from 10% to 12% in the morning, then announced 15%, and in the evening the government withdrew sterling from the mechanism. The next day the rate was back to 10%. The lesson: a promise (‘we will defend the parity’) is worth the cost the government is willing to pay, and the market computes that cost.
Uncovered interest parity, with illustrative (not historical) figures: domestic rate 6.5%, euro rate 2.5%, spot 4.90 lei per euro. The one-year rate implied by parity: 4.90 × 1.065 / 1.025 = 4.90 × 1.03902 = 5.09. Check with 1,000 lei: at the domestic rate they give 1,065 lei; converted into euro, 1,000 / 4.90 = 204.08 euro, which at 2.5% give 209.18 euro, and reconverted at 5.09 give 209.18 × 5.09 ≈ 1,065 lei. Equality means the market expects no difference. If a year later the actual rate is 4.95, the same 209.18 euro are worth 1,035.4 lei, a loss against 1,065 for whoever chose the euro; conversely, whoever borrows in euro gains. When high-interest currencies depreciate less than parity predicts, the carry trade wins and parity is rejected empirically.
For the forecaster: the reference class. For the breaking of a managed or fixed exchange-rate regime, the reference class is built from crisis databases: Laeven and Valencia (IMF) for currency crises 1970–2011, and Reinhart and Rogoff’s ‘This Time Is Different’ for their history; count how many countries with an anchor regime saw a depreciation of more than 15% within a year, relative to the total number of country-years with an anchor. We do not have a verified base number here; it is obtained by counting, with thresholds fixed in advance.
Indicators to watch: the EUR/RON reference rate and international reserves (BNR, monthly releases); the gap between the BNR’s policy rate and the ECB’s (BNR, ECB); the current-account balance and its financing (BNR, see lesson 13); implied volatility and forward quotes, where public (ECB for reference rates).
The trap: taking a managed rate as proof of a fixed regime. The BNR smooths movements but does not guarantee a parity; a long period of stable rates says nothing about the cost of defending it. A second trap: forecasting a point value at 3–12 months with a structural model, when Meese and Rogoff show that ‘today’s rate’ is a hard benchmark to beat.
A scorable question (illustrative): ‘Will the EUR/RON reference rate published by the BNR exceed 5.2000 on at least one business day between 1 January and 31 December 2027?’ Resolution source: the daily reference rate published by the BNR, four decimals; exactly 5.2000 does not count; if the BNR stops publishing the rate, the forecast is void.
Chapter 4Crises and the world system
- 19
Financial crises follow years of expanding credit and ever more fragile funding (Minsky); once panic breaks out, the central bank lends freely, dearly and against good collateral (Bagehot).
Lombard Street: A Description of the Money Market · Walter Bagehot · 1873
Walter Bagehot set out the rule that bears his name in Lombard Street (1873): in a panic, the central bank lends freely, at a very high rate and against good collateral. The high rate, he wrote, acts as “a heavy fine on unreasonable timidity”: it discourages those who ask for money only out of fear. Hyman Minsky explained how we get there: after years of calm, funding slides from the hedge form (income pays interest and principal) to the speculative form (income pays only the interest) and to the Ponzi form (not even the interest). Charles Kindleberger, in Manias, Panics, and Crashes (1978), put the pieces in sequence: shock, credit boom, euphoria, panic. The mechanism has two halves. Before the crisis, credit grows faster than income: Schularick and Taylor showed in 2012, with data for 14 countries from 1870 to 2008, that credit expansion is a reliable signal of crisis risk. In the crisis, assets find no buyers and short-term funding disappears. On 15 September 2008 the investment bank Lehman Brothers, with assets of about 639 billion dollars, filed for bankruptcy protection, the largest bankruptcy in US history. Interbank markets froze, and central banks began supplying liquidity on an unusual scale, which is exactly what Bagehot had prescribed. The limits are real. The rule assumes you can tell an illiquid bank from an insolvent one, and in the middle of a panic that distinction is a judgement call; public support also raises moral hazard. Minsky explains why crises come, not when: fragility can last for years. Two things matter for a forecaster: credit and the structure of funding are tracked directly, not through statements, and a ‘banking crisis’ is a class of rare events with a countable base rate.
First. That these loans should only be made at a very high rate of interest. […] Secondly. That at this rate these advances should be made on all good banking securities […].Walter Bagehot, «Lombard Street: A Description of the Money Market», Henry S. King, Londra, 1873, cap. VII, par. 58–59 (ediția digitală Econlib); româna: traducerea noastră
Why it matters Lehman Brothers failed on 15 September 2008 with assets of about 639 billion dollars, and expanding credit remains the best early signal of a crisis.
In full →Collapse ↑
Worked example: three firms and a rising interest rate. Take three fictional firms, each with debt of 100 million, an annual principal repayment of 10 million and interest at 5%, so 5 million a year; the total obligation is 15 million. Their expected annual incomes differ:
Firm Income Interest Interest+principal Class (Minsky) A 20 5 15 hedge (20 ≥ 15) B 8 5 15 speculative (5 ≤ 8, but 8 below 15) C 3 5 15 Ponzi (3 below 5)Firm B covers the interest but is 7 million short of the total obligation, which it obtains by refinancing; firm C covers not even the interest, and its debt grows by 2 million a year. Now the interest rate rises. At 8%, interest is 8 million and the total obligation 18: firm A (20 ≥ 18) stays hedge, firm B (8 = 8) only just pays its interest, and firm C has a deficit of 5 million a year. At 9% interest is 9 million and B, with income of 8, becomes Ponzi. Firm A changes class only above 10%: at 10%, interest (10) plus principal (10) equals its income of 20; at 11% the obligation is 21 and A becomes speculative.
Minsky’s point shows in the numbers: the same rise in the rate, from 5% to 11%, moves different firms at different moments, without any income having fallen. Each firm that moves to a more fragile class depends more on someone else refinancing its debt. On a real scale, 15 September 2008 showed what happens when many speculative borrowers depend on daily refinancing: Lehman Brothers, with assets of about 639 billion dollars, filed for bankruptcy protection that day, and short-term funding markets froze.
For the forecaster: the base rate. The database of Laeven and Valencia (IMF, 2018 edition) counts 151 systemic banking crises between 1970 and 2017, that is 151 / 48 ≈ 3.1 a year worldwide. The rate per country-year is obtained by dividing the number of crises by the number of country-years in the sample; it is small and is adjusted with the signals below, not replaced by them.
Indicators to follow: the credit-to-GDP gap (Bank for International Settlements, quarterly series for dozens of countries); house prices (BIS, Eurostat); the composite indicator of systemic stress, CISS, of the European Central Bank, daily for the euro area; for Romania, credit to the private sector and the share of non-performing loans in the National Bank of Romania’s financial stability report.
The typical trap: confusing a vulnerability with a date. Credit can grow for years without a crisis, and a good fragility signal is not a calendar signal. The second trap is the definition. Laeven and Valencia date a crisis by the intensity of the policy response (guarantees, nationalisations), so a question asking for a ‘systemic crisis’ must say which list resolves it and what happens if the list is revised.
Example of a scorable question: ‘Will the CISS indicator for the euro area, published by the European Central Bank in its Data Portal, exceed 0.50 on at least one business day between 1 January and 31 December 2027? The value in the series as published on 15 January 2028 is used; later revisions do not count.’ The question is illustrative and gets no probability here: you build one yourself, starting from the base rate of days above the threshold in the historical series.
- 20
A currency crisis does not begin with the exchange rate: it begins with an external deficit financed by short-term foreign-currency debt beyond reserves; when lenders withdraw, adjustment comes within weeks.
The Twin Crises: The Causes of Banking and Balance-of-Payments Problems · Graciela L. Kaminsky, Carmen M. Reinhart · 1999
Guillermo Calvo called a ‘sudden stop’, in 1998, the unexpected interruption of foreign capital inflows that forces a country to close its external deficit in a very short time. Graciela Kaminsky and Carmen Reinhart showed in The Twin Crises (1999), on a sample of countries from the last decades of the twentieth century, that banking problems and balance-of-payments crises often appear together, and that financial liberalisation precedes both: the causes are common, not a simple link from banks to the exchange rate. Barry Eichengreen and Ricardo Hausmann added ‘original sin’: a country that can borrow abroad only in foreign currency risks seeing a depreciation inflate its debt. The mechanism has a breaking point: reserves. The Guidotti–Greenspan rule asks that official reserves cover at least the external debt with residual maturity of up to one year. Thailand let its currency float on 2 July 1997, after gross reserves had fallen from 37.2 to 30.9 billion dollars in six months, while the central bank also held forward contracts of about 23 billion, so net reserves were far smaller. Romania went through a stop in 2008–2009: after Lehman’s failure, the financing of a current-account deficit of more than 10% of GDP withdrew, and the 2009 agreement with the IMF, the EU and other institutions came to about 20 billion euros. The limits: early-warning models give many false alarms, and crises are also self-sustaining, since a feared depreciation happens partly because it is feared. The three checks — reserve cover, external deficit, currency structure of debt — measure how large the damage would be, not when the shock comes; the timing is supplied by external shocks.
Financial liberalization usually predates banking crises; indeed, it helps predict them. Rather than a causal relationship from banking to balance-of-payments crises, the macroeconomic ‘stylized facts’ that characterize these episodes point to common causes.Graciela L. Kaminsky, Carmen M. Reinhart, «The Twin Crises: The Causes of Banking and Balance-of-Payments Problems», Federal Reserve Board, IFDP nr. 544, 1996 (rezumat); American Economic Review 89(3), 1999 (traducerea noastră)
Why it matters In 2009 Romania received a package of about 20 billion euros (12.95 billion from the IMF) after external financing withdrew abruptly.
In full →Collapse ↑
Worked example: Thailand’s gross and net reserves, 1997. Figures quoted in the literature on the crisis: the central bank’s gross foreign-exchange reserves were 37.2 billion dollars in December 1996 and 30.9 billion in June 1997, and the bank’s forward position (dollars sold forward, to be delivered later) reached about 23.4 billion.
Gross reserves, Dec 1996 37.2 bn $ Gross reserves, Jun 1997 30.9 bn $ Fall over six months 6.3 bn $ (6.3 / 37.2 = 16.9%) Forward obligations 23.4 bn $ Net reserves = 30.9 − 23.4 7.5 bn $ Net reserves / gross reserves 24.3% (7.5 / 30.9)The reserve that mattered for defending the exchange rate was therefore a quarter of the one on display. The Guidotti–Greenspan rule is just as simple to apply. Take a fictional country with reserves of 40 billion and external debt with residual maturity of up to one year of 60 billion: cover is 40 / 60 = 66.7%, and the gap to the 100% threshold is 20 billion. A country with cover below 100% does not enter a crisis automatically, but it depends on refinancing the 60 billion every year: if lenders refuse, the 20-billion gap has to be closed by cutting imports, by depreciation or through an external agreement.
Romania in 2009, as an order of magnitude for such a gap: the announced package was about 19.95 billion euros, of which 12.95 billion from the IMF (64.9%), 5 billion from the European Union (25.1%), 1 billion from the World Bank (5.0%) and the remaining 1 billion or so (5.0%) from other international financial institutions. The IMF arrangement was approved by its Executive Board on 4 May 2009.
For the forecaster: the database of Laeven and Valencia (IMF) contains, besides banking crises, dated currency crises and sovereign debt crises, so the reference class can be counted by country and year; without a count, do not give a probability. A new episode is compared with episodes in which the same combination of vulnerabilities existed and no crisis followed, not only with those that ended badly.
Indicators to follow: official reserves against external debt with residual maturity of up to one year (the central bank’s external debt statistics, the World Bank, the BIS); the current-account balance (balance of payments published by the National Bank of Romania and Eurostat); the share of foreign-currency debt in public and private debt (the National Bank, the Ministry of Finance); the 10-year government bond yield and its spread over Germany (the National Bank, Eurostat).
The typical trap: treating vulnerability as a date. A country with weak reserve cover can go for years without a crisis, as long as lenders keep refinancing; the warning model says how large the damage would be, not when the shock starts. The second trap is the headline reserve: check whether part of it is encumbered by forward obligations or illiquid.
Example of a scorable question: ‘Will the IMF announce, in a press release on its website, approval by its Executive Board of a new financing arrangement (stand-by, extended or precautionary line) for Romania between 1 January and 31 December 2027? Resolution: IMF press releases in the period; a mere staff-level agreement does not count.’ The question is illustrative; it gets no probability here.
- 21
The dollar dominates reserves, invoicing and global debt; Triffin’s dilemma explains why the system broke in 1971, and Rey’s global financial cycle why the Fed moves even floating-rate economies.
Dilemma not Trilemma: The Global Financial Cycle and Monetary Policy Independence · Hélène Rey · 2013
At the Bretton Woods conference (July 1944) the world chose a dollar convertible into gold at 35 dollars an ounce, and the other currencies were pegged to the dollar. Robert Triffin showed in Gold and the Dollar Crisis (1960) the contradiction in the system: to supply the world with reserves, the US had to run external deficits, but the deficits increased dollar liabilities relative to the gold stock and weakened confidence in convertibility. Valéry Giscard d’Estaing, when he was France’s finance minister in the 1960s, called the American advantage an ‘exorbitant privilege’: Washington borrowed in its own currency. The dilemma was resolved by a rupture: on 15 August 1971 President Nixon suspended the dollar’s convertibility into gold, and in the following years fixed rates gave way to floating ones. The dollar nevertheless remained the dominant currency. According to the IMF (COFER), at the end of 2024 the dollar made up 57.8% of allocated reserves, and the renminbi 2.18%. Hélène Rey argued in 2013, at Jackson Hole, that a global financial cycle, tied to US monetary policy and to risk appetite, moves capital flows whatever the exchange-rate regime: monetary autonomy requires a managed capital account. The limits: the COFER share also moves through exchange rates, not only through decisions; in the last quarter of 2024 its rise was due, according to the IMF, to the depreciation of the other currencies against the dollar. Others argue that the exchange-rate regime still protects partially. For a forecaster, the share in reserves is a slow variable with valuation effects; it says nothing about invoicing, payments or debt, where the dollar may follow a different path.
[…] independent monetary policies are possible if and only if the capital account is managed, directly or indirectly via macro-prudential policies.Hélène Rey, «Dilemma not Trilemma: The Global Financial Cycle and Monetary Policy Independence», Kansas City Fed, simpozionul Jackson Hole, august 2013; NBER w21162 (traducerea noastră)
Why it matters At the end of 2024 the dollar still held 57.8% of central banks’ allocated reserves (IMF, COFER), so Fed policy reaches every open economy.
In full →Collapse ↑
Worked example: a reserve that ‘grows’ in dollars with no purchase. The IMF’s observation for the last quarter of 2024 can be rebuilt with a simple calculation. A fictional central bank holds reserves of 100, expressed in dollars: 60 in dollars and 40 in euros. The dollar’s share is 60 / 100 = 60%.
Before: dollars 60 + euros 40 = 100 dollar: 60.0% Euro falls 5% against the dollar: dollars 60 + euros 40 × 0.95 = 98 dollar: 60 / 98 = 61.2% To return to 60 / 40: target dollars 0.60 × 98 = 58.8; euros 0.40 × 98 = 39.2 sell 1.2 dollars, buy 1.2 of euros (38 + 1.2 = 39.2)The dollar’s share rose by 1.2 percentage points without the bank having bought a single dollar: the whole effect comes from the exchange rate. The IMF’s COFER reports say the same thing on a large scale: the dollar’s share of allocated reserves rose from about 57.3% in the third quarter of 2024 to 57.80% in the fourth, and the IMF attributed the move to the depreciation of the other currencies against the dollar, not to new decisions by central banks. The figure was later revised to 57.79%.
The lesson: a change of 0.5 points in a currency’s share must not be read as a vote by central banks. The exchange-rate effect has to be separated from the portfolio effect, that is, from what banks actually bought or sold. The IMF also publishes values at constant exchange rates, and a question about ‘diversification’ resolves more cleanly on those than on the share.
For the forecaster: there is no good base rate for ‘change of reserve currency’, because the episodes are too few (sterling, the dollar) and slow; say so openly and stay with slow, quarterly measurable variables. One possible reference class: the annual change in the dollar’s COFER share over the last 25 years, which you compute yourself from the IMF series.
Indicators to follow: the share of currencies in allocated reserves, COFER (IMF, quarterly, with a lag of about three months); foreign holdings of US securities, the TIC report of the US Treasury (monthly); the share of currencies in international payments, the monthly SWIFT tracker for the renminbi; the Fed’s policy rate and the VIX volatility index, as measures of the global financial cycle.
The typical trap: confusing reserves with invoicing, payments and debt. The dollar’s share can fall in reserves and rise in cross-border debt. The second trap is the valuation effect described above; a question about ‘the share at year-end’ moves by itself with the euro exchange rate. State whether you use shares at current or constant exchange rates.
Example of a scorable question: ‘Will the dollar’s share of allocated foreign-exchange reserves, according to COFER data published by the IMF for the fourth quarter of 2026, be below 56.00%? Resolution: the value in the first release of the data for that quarter (around March 2027); later revisions do not count.’ The question is illustrative and gets no probability here; start from the historical series and from what you know about the euro–dollar rate.
- 22
Trade pays through relative, not absolute, cost (Ricardo); and a tariff is not automatically paid by the exporter: in 2018 the US paid almost all of it itself, through import prices.
On the Principles of Political Economy, and Taxation · David Ricardo · 1817
David Ricardo showed in On the Principles of Political Economy, and Taxation (1817, chapter VII) that two countries gain from trade even if one produces everything more cheaply: relative cost matters, not absolute cost. Each specialises in the good for which it gives up the least of the other. Jan Tinbergen added the gravity model in 1962: trade between two countries rises with the size of their economies and falls with the distance between them, a regularity that still holds in the data today. Tariffs strike at this mechanism, and the question is who pays them. The Smoot–Hawley Act, signed on 17 June 1930, raised American import duties on thousands of products, and partners answered with their own duties; economic historians still discuss how much of the collapse of trade in the 1930s is due to the Act and how much to the depression. For the 2018 case, Mary Amiti, Stephen Redding and David Weinstein (Journal of Economic Perspectives, 2019) found complete pass-through of American tariffs into import prices: foreign exporters did not lower their prices, and the cost fell on American buyers. In the working-paper version of the study, the loss of real income was about 1.4 billion dollars a month by the end of 2018. The limits: Ricardo’s model has a single factor of production and fixed costs; it ignores job adjustment and modern supply chains, in which a part crosses the border several times and the tariff accumulates. For a forecaster, the announced tariff is not the applied tariff: the date of entry into force, the exemptions and what happens to imports ahead of the deadline all matter.
England may be so circumstanced, that to produce the cloth may require the labour of 100 men for one year; and if she attempted to make the wine, it might require the labour of 120 men for the same time.David Ricardo, «On the Principles of Political Economy, and Taxation», John Murray, Londra, 1817, cap. VII «On Foreign Trade» (text Econlib); româna: traducerea noastră
Why it matters In 2018 American tariffs were passed through completely into import prices, so the cost fell on buyers in the US, not on foreign exporters.
In full →Collapse ↑
Worked example: the gain from specialisation, with Ricardo’s numbers. Costs in man-years, those of chapter VII of the 1817 work:
Cloth (man-years) Wine (man-years) England 100 120 Portugal 90 80Portugal produces both goods with less labour, so it has an absolute advantage in both. Opportunity costs, however, show something else. In England 1 cloth costs 100 / 120 = 0.833 wine; in Portugal 1 cloth costs 90 / 80 = 1.125 wine. Cloth is cheaper, as a sacrifice of wine, in England; conversely, wine is cheaper in Portugal.
Without trade, each country makes 1 cloth and 1 wine: England uses 100 + 120 = 220 man-years, Portugal 90 + 80 = 170. Together: 2 cloth and 2 wine. With trade and the same labour, England makes only cloth: 220 / 100 = 2.2 cloth; Portugal only wine: 170 / 80 = 2.125 wine. Total output rises to 2.2 cloth and 2.125 wine, that is 0.2 cloth and 0.125 wine more.
Exchange takes place at a price between the two opportunity costs, for example 1 cloth for 1 wine (between 0.833 and 1.125). England sells 1.1 cloth and receives 1.1 wine: it ends with 1.1 cloth and 1.1 wine, against 1 and 1 before. Portugal sells 1.1 wine and receives 1.1 cloth: it ends with 1.1 cloth and 2.125 − 1.1 = 1.025 wine, against 1 and 1. The gains add up exactly to the total surplus: 0.1 + 0.1 = 0.2 cloth and 0.1 + 0.025 = 0.125 wine. Comparative advantage does not require anyone to be better at anything in absolute terms; it requires only that relative costs differ.
For the forecaster: a reference class for ‘announced tariffs that enter into force’ can be built from official notices (the Federal Register, USTR, the Official Journal of the EU): you count announcements, entry-into-force dates, the rate at entry and the rate after six months; without this counted table, give no probabilities. Think about the distance between announcement and application, not only about the size of the tariff.
Indicators to follow: tariff notices and exemption lists (Federal Register, USTR; the European Commission for the EU); monthly bilateral trade (the US Census Bureau, Eurostat Comext); the import price index (the Bureau of Labor Statistics); the World Trade Organization’s report on restrictive measures.
The typical trap: confusing the statutory rate with the effective rate. Exemptions, trade diversion through third countries and imports brought forward ahead of the deadline make the effective average tariff differ greatly from the announced one. The second trap is assuming the exporter pays; in the 2018 case pass-through into import prices was complete.
Example of a scorable question: ‘Will China’s share of US goods imports, according to the Census Bureau’s annual data for 2027, be below 10.0%? Resolution: the value in the country trade table published by 31 March 2028; later revisions do not count.’ The question is illustrative and gets no probability here; build one from the historical share series and from the calendar of tariffs already applied.
Chapter 5The forecaster's synthesis
- 23
The macro consensus does not see recessions: they appear in data late, forecasts are revised slowly, they are dated months later, and relations estimated on the past break when policy changes.
How accurate are private sector forecasts? Cross-country evidence from consensus forecasts of output growth · Prakash Loungani · 2001
In 2001 Prakash Loungani measured (International Journal of Forecasting) consensus forecasts of GDP growth between 1989 and 1998 and concluded: ‘the record of failure to predict recessions is virtually unblemished’. Zidong An, João Tovar Jalles and Loungani took up the question again in 2018, for 63 countries and the years 1992–2014: forecasters know that recession years are different, but they get the size of the fall badly wrong until close to the end of the year, forecasts are revised slowly, and private and official forecasts are virtually identical. Two critiques explain why this is not mere negligence. Robert Lucas showed in 1976 that relations estimated on the past change when the policy rule changes, because people adapt; Charles Goodhart observed in 1975 that a statistical regularity used for control loses its link to what it measured. Data add to this: a recession is seen late. The NBER committee announced the December 2007 peak only on 1 December 2008, and the March 2001 peak on 26 November 2001, that is, after almost a year and after eight months respectively. The conclusion is not that macroeconomics cannot be forecast, but that turning points are the hardest to catch and the consensus is a weak starting point. A forecaster uses it as an anchor, not as a probability, and asks questions about figures published on a fixed date, not about turning points: there the Brier score quickly shows whether you added anything beyond the consensus.
The record of failure to predict recessions is virtually unblemished.Prakash Loungani, «How accurate are private sector forecasts? Cross-country evidence from consensus forecasts of output growth», International Journal of Forecasting 17(3), 2001, pp. 419–432, rezumat (traducerea noastră)
Why it matters Loungani found that in 1989–1998 consensus forecasts missed practically all recessions, and the NBER announced the December 2007 peak only after almost a year.
In full →Collapse ↑
Worked example: why catching a few recessions does not beat a flat forecast. The numbers are illustrative, built to show the arithmetic, not real data. In a sample of 200 country-years, 20 are recessions: the base rate is 10%. We compare two forecasters with the one-sided Brier score, (p − outcome)², averaged over cases.
The constant forecaster says p = 0.10 everywhere. For the 20 recessions, (0.10 − 1)² = 0.81; for the 180 years without recession, (0.10 − 0)² = 0.01.
Constant (p = 0.10): 20 × 0.81 = 16.20 180 × 0.01 = 1.80 total 18.00 → 18.00 / 200 = 0.0900The alarmist forecaster gives p = 0.8 in 8 country-years, of which 3 are recessions and 5 are not, and p = 0.05 in the remaining 192, of which 17 are recessions and 175 are not. It caught 3 of 20 recessions, that is 15%, with an alarm precision of 3 / 8 = 37.5%.
Alarmist: 3 × (0.8 − 1)² = 3 × 0.04 = 0.12 5 × (0.8 − 0)² = 5 × 0.64 = 3.20 17 × (0.05 − 1)² = 17 × 0.9025 = 15.3425 175 × (0.05 − 0)² = 175 × 0.0025 = 0.4375 total 19.10 → 19.10 / 200 = 0.0955The forecaster who ‘sees’ recessions has a worse score (0.0955) than the one who sees nothing (0.0900). Against the flat version: the 3 recessions caught gain 3 × (0.81 − 0.04) = 2.31, but the 5 false alarms lose 5 × (0.64 − 0.01) = 3.15, and the 17 recessions missed at p = 0.05 lose 17 × (0.9025 − 0.81) = 1.5725; the 175 quiet years gain 175 × (0.01 − 0.0025) = 1.3125. Net: −2.31 + 3.15 + 1.5725 − 1.3125 = +1.10, that is 19.10 − 18.00. A signal that catches 15% of the turning points, with 37.5% precision, is not enough. That is why a consensus that misses turning points, as Loungani (2001) and An, Jalles and Loungani (2018) describe it on real data, adds little beyond the base rate precisely on recessions.
For the forecaster: a simple annual definition, used in the literature, calls a recession a year with negative real GDP growth; you can count by country and year from the IMF World Economic Outlook series or from national accounts (the National Institute of Statistics for Romania, Eurostat for the EU). The base rate that results is a better starting point than the consensus, especially for questions about turning points.
Indicators to follow: IMF forecasts in the World Economic Outlook (April and October) and those of the European Commission, with their history; the flash GDP estimate (the National Institute of Statistics about 45 days after the quarter, Eurostat about 30 days); the first release against the revised series for the same period; NBER dating, for the US.
The typical trap: scoring on revised data when the question should have been resolved on the first release, or the other way round. Revisions can move a small growth figure from one side of zero to the other. The second trap is treating the consensus as a probability; it is a central point, not a distribution, and in turning-point years it lies outside the range.
Example of a scorable question: ‘Will the flash estimate of Romania’s GDP growth in the second quarter of 2027, quarter on quarter, seasonally adjusted, be negative? Resolution: the National Institute of Statistics release published by 15 August 2027; later revisions do not count.’ The question is illustrative and gets no probability here; start from the base rate of negative-growth quarters in the INS series and then adjust with the data already published for the period.
- 24
An analysis becomes a forecast when it passes through a chain: signal, mechanism, counted reference class, question with deadline and source, update; only then can it be scored.
Superforecasting: The Art and Science of Prediction · Philip E. Tetlock, Dan Gardner · 2015
An analysis becomes a scorable forecast through a chain: signal, mechanism, reference class, question with deadline and source, update. Philip Tetlock and Dan Gardner, in Superforecasting (2015), describe the same moves: triage of questions, breaking them into parts, starting from the base rate (the outside view) and only then adjusting with specific information (the inside view). Nothing in the domain knowledge of this course comes into play without the last link: the question a stranger can resolve. The reference class has to be counted, not narrated. The NBER committee dated 12 American recessions with peaks between November 1948 and February 2020, so 12 starts in 73 calendar years (1948–2020): about 16% a year. Other classes are built the same way: interstate wars from the Correlates of War, banking, currency and debt crises from the IMF’s Laeven–Valencia database, published central-bank decisions at each meeting. Then the signal updates the class: if an indicator appears in 90% of the months before recessions and in 20% of the others (illustrative figures), the likelihood ratio is 4.5, and a base rate of 16% becomes, by Bayes’ rule, about 47%. The limits: 12 cases are a small sample, and an indicator that caught 8 of 12 is still an uncertain estimate; a class chosen after the outcome or an inflated likelihood ratio gives false certainty. For calibration and scoring, the Superforecasting course remains the framework; what is new here is placing your domain knowledge in a worksheet.
Triage. Focus on questions where your hard work is likely to pay off. […] Break seemingly intractable problems into tractable sub-problems.Philip E. Tetlock, Dan Gardner, «Superforecasting: The Art and Science of Prediction», Crown, 2015, anexa «Ten Commandments for Aspiring Superforecasters» (citat după rezumate secundare; traducerea noastră)
Why it matters The NBER dated only 12 American recessions between 1948 and 2020, so the base rate of about 16% a year can be counted in a few minutes.
In full →Collapse ↑
Worked example: two worksheets and a computed update. The macro worksheet for ‘recession in the US’, with figures worked step by step:
SIGNAL warning indicator (for example, the yield curve, lesson 16) MECHANISM dearer credit → weaker investment and consumption → fall in activity CLASS 12 NBER recession starts in 73 years (1948–2020) BASE RATE 12 / 73 = 0.164 → odds 12 : 61 = 0.1967 LIKELIHOOD (illustrative) P(signal | month before a recession) = 0.9 P(signal | other month) = 0.2 → ratio 0.9 / 0.2 = 4.5 UPDATE new odds = 0.1967 × 4.5 = 0.885 (54 : 61) p = 54 / (61 + 54) = 54 / 115 = 0.47 QUESTION recession begins in window X, resolved by NBER datingThe figures 0.9 and 0.2 are invented to show the arithmetic; in a real analysis you count them from history, keeping the small sample in mind: with only 12 recessions, a change of one case moves the ratio a lot. The base rate alone (16%) gives one probability; the signal moves it towards 47%, and distrust of the signal’s figures can pull it back towards the middle.
The geopolitical worksheet follows the same scheme with different pieces. Signal: a concentration of forces at the border, with visible costs that are hard to withdraw (lesson 5). Mechanism: a commitment problem or private information (Fearon). Reference class: interstate wars in the Correlates of War, counted by pairs of states and years, with the threshold of 1,000 battle deaths. Question: what event, what threshold, what deadline, what source. Update: each jump in the signal’s costs or each offer to negotiate moves the likelihood ratio. The worksheet is complete only if every line has a figure or a source; a line without them stays a hypothesis.
For the forecaster: easily counted reference classes. US recessions (the NBER list, 12 starts between November 1948 and February 2020). Interstate wars (Correlates of War, threshold of 1,000 battle deaths). Banking, currency and debt crises (the Laeven–Valencia database, IMF). Central-bank decisions: at each meeting the rate changes or not; you count the meetings in the published archive of the National Bank of Romania, the ECB or the Fed and compute the frequency of changes, then condition it on the state of the economy.
Indicators to follow, one for each link of the chain: the official calendar and releases of the central bank (the National Bank of Romania, the ECB, the Fed); the series at the resolution source (the National Institute of Statistics, Eurostat, the Census Bureau); the database used as the class (NBER, COW, IMF); your own forecast log, with the date, the probability and the reason for each update.
The typical trap: the convincing story. A coherent causal chain seems more probable than the product of the probabilities of its links; each extra link makes it less probable, not more solid. The second trap: a class chosen after the outcome (Allison’s 12 of 16 cases) or a likelihood ratio taken from memory.
Example of a scorable question: ‘Will the National Bank of Romania announce, in the release after the first monetary-policy meeting of 2027, an unchanged policy rate? Resolution: the release published on the National Bank’s website on the day of the meeting; if the meeting is moved, the first one in 2027 is used.’ The question is illustrative and gets no probability here. For scoring, calibration and the forecast log, the Superforecasting course is the natural continuation.
How to cite this page
Plain
olivLaw, «Geopolitics and macroeconomics for forecasters», https://olivlaw.com/en/eu/aprofundeaza/geopolitica-si-macroeconomie (accessed November 20, 2026).
APA
olivLaw. (n.d.). Geopolitics and macroeconomics for forecasters. https://olivlaw.com/en/eu/aprofundeaza/geopolitica-si-macroeconomie
BibTeX
@misc{olivlaw-nd-geopolitica-si-macroeconomie,
author = {{olivLaw}},
title = {Geopolitics and macroeconomics for forecasters},
howpublished = {\url{https://olivlaw.com/en/eu/aprofundeaza/geopolitica-si-macroeconomie}},
note = {Accessed 2026-11-20},
}The texts are produced by olivLaw's AI pipeline; the author of record is the organisation.