Geopolitics is the study of how states, and actors who are not states, shape the terms on which economic activity happens across borders. The reason it belongs in an analyst's toolkit is narrow: it arrives at a company through four channels and only four. A cost channel, when an input is taxed, restricted or re-routed. A revenue channel, when a customer or a market becomes unreachable. A licence channel, when permission to operate, list, own or transfer is granted or withdrawn. An asset channel, when property is seized, stranded or destroyed.
Every geopolitical event that has ever mattered to a valuation came in through one of those four doors.
Who are the actors, and how do states relate?
The actors divide in two. State actors hold territory, a legal monopoly on force, and the power to write and enforce rules; they act through governments, central banks, sovereign funds and state-owned enterprises. Non-state actors hold influence without territory: multinational firms, standard-setting bodies, armed groups, and the operators of infrastructure everyone depends on.
Advertisement
A payments network that can exclude a country's banks is exercising a power that used to require a navy.
The relationships between states fall into four archetypes, and naming the one you are in disciplines a forecast.
Autarky is self-sufficiency by design: a state minimises its dependence on others, accepting a lower standard of living for a higher degree of control. Hegemony is a single dominant state setting the rules, providing the reserve currency and the security guarantee, and bearing the cost of both. Bipolarity is two roughly matched blocs, each internally cooperative and mutually antagonistic, with third countries forced or courted into alignment. Multipolarity is several centres of power with shifting alignments, cooperation issue by issue rather than bloc by bloc, and no single guarantor of anything.
Cooperation is not a moral variable here. It is a function of standardisation and of the cost of exit: states cooperate where the gain from a common standard is large and the cost of defecting is high, which is why aviation safety and payments messaging survive political hostility that trade in advanced semiconductors does not.
Tools of geopolitics, and the limit of forecasting
The tools of geopolitics come in three families. National-security tools, which are military posture, alliances, basing, intelligence and territorial control. Economic tools, which are tariffs, quotas, subsidies, industrial policy, export controls, investment screening and procurement preference. Financial tools, which are sanctions on entities and individuals, exclusion from payment and messaging systems, restrictions on capital flows, asset freezes and the weaponisation of a reserve currency's plumbing. The financial family has grown fastest and moves fastest, because it requires no logistics: a designation takes effect the day it is published.
One honest limit before you use any of this. Geopolitical forecasting has a poor and well-measured track record, and an analyst's edge is almost never in predicting the event. It is in knowing beforehand exactly which line of which model moves and by how much.
When the headline arrives, you should be repricing rather than reading.
Classifying and measuring geopolitical risk
A classification is only worth having if it changes what you do, and this one does, because each class implies a different instrument.
An event risk is discrete and dateable: an election, a treaty deadline, a sanctions designation, a court ruling. You know when to look. The instrument is a scenario with a date attached, and often an option, since the timing is knowable even when the outcome is not.
An exogenous risk arrives without warning and reorders the environment: an invasion, a coup, a pandemic, a disaster closing a strait. You cannot date it. The instrument is resilience rather than prediction, which in a model means asking what breaks and how fast it can be repaired.
A thematic risk is a slow, directional shift visible for years before it binds: supply-chain reconfiguration, carbon border pricing, export controls tightening ratchet by ratchet. You can see it coming and still be surprised by the timing. The instrument is a re-underwriting of the long-run assumptions.
Thematic risk is where most of the money actually is, precisely because nothing forces you to act on any given day.
How do you turn a classification into a number?
Three attributes turn a classification into a number.
Velocity is how fast the effect arrives once the trigger fires. A sanctions designation has a velocity of days. A carbon border levy phased in over five years has a velocity of years.
Velocity decides whether you can act at all after the event, and it is the attribute most often left out of a register.
Likelihood is a probability you are obliged to write down. A register with likelihoods stated as "high" and "medium" cannot be multiplied by anything, and therefore cannot be checked against outcomes, which means it can never be improved.
Impact is a currency amount on a named line, not a rating.
"Margin pressure" is not an impact. "₹95.04 crore of contribution lost, being 9% of revenue at a 22% contribution margin" is an impact, and it can be wrong in a way you will notice.
Geopolitical risk indices and exposure maps
Published indices exist and are worth knowing. The Geopolitical Risk index of Dario Caldara and Matteo Iacoviello counts newspaper mentions of adverse geopolitical events and is free with a long history; the Economic Policy Uncertainty index of Scott Baker, Nicholas Bloom and Steven Davis does the same for policy uncertainty. Both are text-based counts, which is their strength and their limitation. They are reproducible over decades, and blind to a risk nobody has written about yet.
Use them to date and size a regime, never to find the risk.
A firm-level exposure map beats any index for a specific company, and it is built from the filing rather than from the news. Four lines, each a percentage you can source: revenue by destination and end-market, cost by origin and single-source dependency, licences by issuing authority and expiry, and assets by physical location. Worked Example 8 builds that map and converts it into an expected annual drag.
{
"skill": "Turning a filing into a firm-level geopolitical exposure map with numbers on every row",
"use": "Use an assistant to pull the geographic revenue split, the raw-material and single-source disclosures, the plant locations and the licence or approval list out of an annual report, and to draft the register rows; then you assign every likelihood, every velocity and every impact yourself and run the expected-value arithmetic.",
"tools": ["Document AI grounded on the annual report or 10-K you already hold", "Chat assistants with code execution for the expected-value and discount-rate arithmetic", "Primary sources: the company's own filing, the issuing regulator's register, and the sanctions or tariff list itself"],
"prompt": "From this annual report, extract into a table: revenue by geography with the page number for each figure, revenue by end-market segment, any disclosed single-source or sole-supplier dependency, the location of every manufacturing site, and every licence, approval or registration the company says it holds with its issuing authority. Then list, separately, every figure I asked for that the report does NOT disclose. Do not infer or estimate any missing number. Quote the exact sentence for any dependency disclosure.",
"verify": "Read the list of what the report does not disclose first, because that list is where the unpriced risk lives, and a model asked for a table will quietly fill a gap rather than report it. Trace each extracted percentage back to its page yourself: document extraction transposes columns in dense segment tables and a transposed geography split is invisible downstream. Confirm every regulatory name against the issuing authority's own register rather than against the model's recollection, since agency names, thresholds and list contents change by notification. Recompute the expected drag by hand for the two largest rows. State the recovery assumption, the pass-through assumption and the duration assumption on every row, because an impact quoted without those three is not checkable.",
"diy": "You must be able to build the four-line exposure map from a filing on your own, assign a probability you would defend, and carry the arithmetic through to a value effect and its discount-rate equivalent, saying which of the two you would actually put in the model. Worked Example 8 is the drill. A tool can find the disclosures; the judgment that a 12% probability is right, and that this risk is idiosyncratic and therefore belongs in the cash flows rather than the discount rate, is not delegable.",
"market": "IN"
}