How a Geopolitical Analyst Predicts the Outcome of War | Odd Lots
What this covers
Andrew Bishop, global head of policy research at Signum Global, joins Joe Weisenthal to explain how geopolitical analysts forecast conflicts and their market consequences—not by predicting single outcomes, but by assembling rigorous processes that decompose chaos into manageable sub-probabilities. The conversation centers on the Israel-Iran conflict, US policy under Trump, and the mechanics of what makes analysis useful to traders betting on oil, grain, currencies, and travel stocks when geopolitics shifts. Bishop argues that the scarcity of historical data makes statistical modeling impossible; instead, analysts must break complex scenarios into independent variables, assign probabilities to each, and compound them to identify tail risks that emerge not from any single likely scenario but from the aggregate of many low-probability paths.
The episode ranges across how leaders actually behave (the "logic of political survival" framework, with limits), why grand theories fail in the short term despite holding over years, and how to diagnose what went wrong when predictions miss—as Bishop did when underweighting Trump's appetite to intervene militarily in Iran. It covers the patterns that fade once they become famous (like Trump's track record of backing down), the risks of proximity to power (Treasury Secretaries don't know what Trump will do next), and why prediction markets, despite their appeal, often constrain analysis by fixing dates that don't match the actual decision points. The conversation also touches Taiwan, the underappreciated threat of a blockade over invasion, and why the collapse in oil prices when the war ended was actually predictable if you focused on timing peak fear rather than war duration. Throughout, Bishop emphasizes that clients value the analysis primarily as scenario rehearsal—like memorizing a race track before driving it—rather than as a single forecast.
Bishop argues that the value of geopolitical risk analysis lies not in accurately predicting outcomes but in rigorous, nimble process — breaking complex situations into compounding sub-probabilities and pattern-recognition — because the data is too sparse and the actors too unpredictable to model precisely.
- Geopolitical events have too few case studies to model statistically, so process beats prediction
- Clients use the analysis to pre-rehearse scenarios (the 'F1 driver' model), not to obtain a single forecast
- Compounding independent probabilities can push an outcome past 50% even when no single scenario is the modal one
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The 'logic of political survival' framework explains leader behavior — leaders place their own power and advantage ahead of everything, which is why Netanyahu would never agree to a hostage deal that risks his coalition just to satisfy Trump — but the framework is not foolproof, as shown by Hamas surrendering hostages twice despite the leverage logic predicting otherwise.
“the logic of political survival right like leaders will always put their own advantage ahead of anything”
Assigning probabilities to sub-events before publishing serves as an internal logic check: if the cumulative math doesn't add up to 100, it signals you are overstating or understating some aspect of the dynamic, even though the value is logical rather than mathematical.
“if the math doesn't add up, that can highlight a problem in your logic”
An overarching grand theory like 'Trump's trade policy is all about China' can be accurate over four years yet useless in the short term, because for the first three months the administration was obsessed with Mexico and Canada — so analysts should avoid grand theories and assess each prediction individually (fox vs hedgehog).
“that is totally accurate over the course of four years it is totally useless in the short term”
A base case scenario can be market-constructive while the aggregate of all non-base-case scenarios produces a more likely market-negative outcome — clustering non-base scenarios revealed this in a Uruguay election analysis.
“if you aggregated all the non-base case scenarios you actually had a more likely outcome of a market negative bottom line”
When a prediction is wrong, the productive response is to diagnose why — in the Iran case Bishop was too focused on Israeli military capabilities and not enough on Trump's political appetite to intervene — yielding a lesson that there is no single 'one rule to rule them all' and analysts must constantly reassess.
“you basically have to try to figure out why you got it wrong, right? And from there you can typically learn something for the next time”
Because there are many independent ways to reach a given outcome, compounding the probabilities of independent variables can tip the cumulative odds past 50% even when no single scenario is your modal (most likely) scenario — for example, US intervention in Iran exceeded 50% overall even with a low probability assigned to Trump wanting to intervene.
“because there are so many ways of getting to an outcome, you can get tipped over 50 without it being your your modal scenario”
A Chinese move against Taiwan within five years is expected, but it is more likely to be a blockade than an amphibious invasion; the underappreciated risks are that a blockade could be rolled out literally overnight with no buildup, and that a sustained bloodless 'anaconda squeeze' blockade of 3-6 months could be far worse for markets than a one-month war regardless of outcome, because of the prolonged disruption to business operations.
“it's more likely to be a blockade for example than an actual amphibious invasion”
Intel-based geopolitical analysis (insider information from Congress, golf with officials, etc.) does not make for good predictions because even insiders like the Treasury Secretary did not know whether Trump would back down on tariffs, and Assad did not know if he would be toppled — proximity to power does not confer foresight.
“it doesn't make for good predictions, right? Because Scott Besson himself didn't know on April 8 whether Trump was going to back down the next day”
True black swans almost never happen; the events that blow up are usually visible well in advance (e.g., months of Russian buildup before invading Ukraine, pandemics ranked top-5 in WEF risk reports for 15 years before COVID) — the real difficulty is predicting the details (timing, location) and knowing how to trade them, not anticipating the event category.
“the whole black swan thing is irrelevant. Like black swans almost never happen. The stuff that blows up in our face is stuff that is visible pretty far ahead”
Predicting geopolitical outcomes is actually harder than flipping a coin or throwing darts, because most situations have far more than two possible scenarios, making accurate probability assignment extremely difficult.
“It's actually harder than just throwing darts or flipping a coin because most situations have far more than two scenarios”
Studies showing geopolitical risk doesn't move markets are short-sighted because they typically measure the S&P over the long term, whereas the firm's clients trade short-term and in niche assets (travel/tourism, oil and gas, Israeli shekel) where geopolitical events produce major sharp volatility.
“most of these studies typically look at S&P in the long term. And our clients obviously are not trading the S&P in the long term”
Trump's pattern of backing down (the 'taco' phenomenon) — backing down in roughly 21 of 23 documented threats — was a valid pattern, but it was already fading by the time it became a popular label, illustrating an 'Economist cover' effect where the pattern becomes less true once widely recognized, requiring constant reassessment.
“I think Trump had backed down in 21 out of 23 threats or something like that”
Geopolitical regime changes (e.g., shift to a multipolar world, US abandoning the hegemon role) do happen but unfold much more slowly than financial-market regime changes, giving more time to price them in; even the isolationist, pacifist-leaning Trump intervening militarily in Iran demonstrates how slow that process is.
“the regime changes happen but they happen much more slowly and so you have a lot more time to bake them in”
It is better to have rigorous analysis with a wrong call than a correct call with bad analysis, because good process is repeatable while a lucky correct call without process is just throwing darts.
“even if you get the call wrong, it's better to have good analysis and a call wrong rather than the other way around”
Identifying Trump's motivation in each specific case is key to prediction: if he is after tariff revenue he cannot back down (you can't collect revenue if you retreat), whereas transactional motivations like fentanyl or immigration concerns with Mexico made him likely to back down once he could claim he had acted — and the same country can be targeted by completely different motivations at different times.
“if you think he's after tariff revenue then he's by definition not going to back down. You can't get revenue if you back down”
Prediction markets are currently of limited use to clients mainly because of low liquidity, and they are more useful to the analyst than to the end user because their rigidity — fixing predictions to specific falsifiable dates — narrows their value when the truly relevant cutoff (e.g., a Trump visit) differs from the market's stated date.
“the rigidity of the prediction markets and it's a trade-off, right? Because the the reason they do them that way is so that they can be easily falsifiable”
The collapse of oil prices when the Israel-Iran war ended was a counterintuitive outcome that many clients actually anticipated, because the relevant question was timing the peak fear level — likely when Iran retaliates for whatever the US does — so prices dropping on the news of the war's end was not surprising to savvy clients.
“what they were trying to figure out was when is this going to end or not end, but when is it going to culminate, right? When is going to be the peak fear level?”
A useful lesson early in an analyst's career is to not get overly excited by every new headline, citing the perennially overhyped 'death of the dollar' stories (first gas contract settled in euros) and the BRICS narrative, which has promised a 'glorious future' for nearly 20 years without materializing.
“the zero hedge type headline about how the first gas contract was denominated or settled in euro instead of dollar... the bricks have had a glorious future for you know almost 20 years”
Rapidly changing geopolitics makes analysis harder because it shrinks the already-tiny historical data set — e.g., the long-established 'axis of convenience' read on the Russia-China relationship became genuinely unclear after the 2022 Ukraine invasion — and the temptation to 'make it up as you go' is where analysis gets dangerous.
“so much stuff is changing so fast means that you have even less historical background or historical data”
Trump allows his advisers to say anything, including the craziest statements, as long as they do not reduce his optionality — he benefits from them going to extremes provided he doesn't get cornered by their comments, which is why parsing his spokespeople literally (e.g., 'sign the executive order' vs 'implement the tariffs') can be predictively useful.
“they are allowed to say anything, including the craziest stuff, as long as they don't reduce his optionality”
Clients value the analysis primarily not for the outcome prediction itself (for which they could use prediction markets) but for the rehearsal of all possible scenarios in advance — like an F1 driver memorizing every turn of the track before a race so they are pre-prepared to react when events unfold.
“they're an F1 driver, and we're helping them on the sort of, you know, the training tracks”
Technical knowledge (e.g., missile ranges, weapons capabilities) is necessary for geopolitical forecasting but need not be held in-house: the firm breaks a big issue into sub-questions and sources answers from open source or external experts while never outsourcing the actual analysis or predictions.
“you do need it, but you can break it down and not make it part of your core business model”
Bishop's final June 16th takeaways concluded that the overwhelming majority of scenarios end in a negotiated Iranian capitulation with no meaningful damage to global oil supply or Gulf assets, which proved a useful market-relevant conclusion even though US intervention (assigned 25%) did occur.
“The overwhelming majority of scenarios end in a negotiated Iran capitulation with no meaningful damage done to global oil supply or Gulf assets”
China may have a price at which it would cut off drone-component supplies to Russia, but that price is so far beyond what the US would be willing to pay (e.g., lifting export controls) that the question is moot in practice.
“even if there were a price, we are so far from being willing to pay that price that it's not going to it's never going to see the light of day”
I always say what I'm concerned about is geopolitical tension — it's a cliché people fall back on when things seem fine.
“what are you concerned about? It's like, I'm concerned about geopolitical tension. It's just one of these clich”
7030 is the new 6040 — analysts who want to sound confident but are unsure now state 70% odds instead of the old 60%.
“it used to be that if you were unsure but wanted to sound smart, you say 6040. But these days you say 70. 7030 is the new 6040.”