Supply Risk Early Warning

Why this matters more now

Commodity prices are getting harder to read, and price charts only show the trouble once it has arrived.

Price swings are becoming larger and more frequent, for two different reasons. This model builds both in, and makes them more frequent over the 30 years it simulates.

Production shocksMore frequent and more severe weather in growing regions, plus local politics. They build over weeks and follow the seasons.

Route shocksShipping chokepoints disrupted, as the Red Sea has been. A route is open one week and closed the next.

Four commodities, shown the usual way

Prices first, as most tools show them. Each commodity has a made-up price history, a handful of growing regions and the shipping routes it uses. They're chosen for contrast: one region dominant (cocoa), two regions dominant (palm oil), spread out (wheat), and spread out but dependent on one chokepoint (sugar).

Price, last three years

production shockroute shockno supply cause

Shocks are shaded with hindsight: a rise of 8% or more within 16 weeks. Each shows in the price only once it has started.

Where it's grown

How it's shipped

What a price-only view says today

The next step adds what the price chart leaves out: how risky the growing regions and the shipping routes are right now.

Add the supply risk

Where the risk sits, before the price moves

Two kinds of risk, kept apart. Growing regions affect how much of a commodity exists. Shipping routes affect how, how fast and at what cost it reaches buyers. Both add up to the price pressure expected over the next 12 weeks.

Growing regions: production risk

Region	Share of world supply	Weather	Politics	Price pressure

lowraisedhigh

Shipping routes: transit risk

Route	Share of traded volume	State	Price pressure

opencongesteddisrupted

Price and price pressure, three years

priceproduction pressuretransit pressurealert line

Which commodities depend on which chokepoints

Share of each commodity's traded volume passing each chokepoint (a cargo can pass two). One disruption hits every commodity that depends on that route, in proportion.

What the model sees

Readings from noisy signals, as forecasts and news are. The true state of each region and route drives the simulated shocks and is hidden from every approach, and from the agent.

Next: three ways to raise an alert, run on the same history.

Compare the approaches

Three ways to raise an alert

The same history, read three ways. Price-only is the usual status quo. The middle one uses where supply is grown and the usual risky season, but no live news, to show how much the live signals add.

Price only

Today's usual tool. Alerts when the price has risen fast over the last four weeks, on a threshold set in calmer years.

Sees: Prices
Warns early?: Rarely: it reacts once the price has moved
As swings grow: Fires more often on noise, and can't tell

Concentration and calendar

Alerts on the most concentrated commodities ahead of their risky season, with no live news.

Sees: Where supply is grown, the time of year
Warns early?: Sometimes, for production shocks
Route shocks: Can't see them

Concentration and live risk

Alerts when expected price pressure from growing regions and shipping routes passes a threshold.

Sees: Where supply is grown and shipped, and live signals for both
Warns early?: Yes, when the signals pick the trouble up
As swings grow: Reads current conditions, not a fixed price rule

Time on alert

Every approach gets the same weeks on alert a year per commodity, set on the first 10 simulated years and then left alone, as a real team would. None can do better just by staying on longer.

Signal quality

How well the regional and route signals track the true state. At "None" they are pure noise, which checks the advantage comes from the signals and not from the setup.

One history, three sets of alerts

Bars show when each approach was on alert. Shaded: shocks, with hindsight. These settings also change the backtest and results tabs.

One history can be lucky. The backtest replays hundreds.

See the backtest

Replay it hundreds of times, in a world getting noisier

Shocks a year, per commodity, across the simulated years

production shocksroute shocksno supply cause

Average over every run. Both supply rates drift upward: production shocks with the climate, route shocks with more frequent chokepoint disruption.

Weeks of warning, every shock caught

price onlyconcentration and calendarlive risk, production shockslive risk, route shocks

Share of each approach's caught shocks by weeks of warning. Negative weeks: the alert came after the price had started to move. Route shocks give less warning: they're sudden, and the lead comes from congestion beforehand and the weeks before delayed cargoes reach the price.

Next: what the runs add up to.

See the results

Earlier warning, and what it costs in false alarms

The three approaches, years 11–20 and 21–30

Shocks warned of 2+ weeks ahead, by simulated year

price onlyconcentration and calendarlive risk

False alarms a year, by simulated year

price onlyconcentration and calendarlive risk

By commodity, live risk

What it would have meant: a Red Sea disruption hits sugar

Next: the same intelligence, asked for by an AI agent instead of read off a dashboard.

Ask the agent

Ask it in plain words

The model's price, production and route data is offered as tools an AI agent can call (an MCP server), so a buyer can ask in their own AI assistant and get an answer built from the same numbers, with the tool calls shown and every figure checked against them.

Ask your own question

Or see an example

Tools the agent can call

list_commodities()

The four commodities with today's price, status and price pressure.

get_price_history(commodity, weeks)

Weekly prices.

get_supply_concentration(commodity)

Growing regions, their shares and current risk, and the concentration index.

get_route_exposure(commodity)

The shipping routes it depends on, how heavily, and their state.

get_route_status(route)

A route's state, for how long, and every commodity that depends on it.

get_exposure(commodity)

Price pressure over the next 12 weeks, split into production and transit, with the drivers ranked.

assess_sourcing(commodity, mix)

Your own exposure, given where you buy from, against the world average.

get_track_record(commodity)

How the alerts did in the backtest, so the agent can say how far to trust one.

Connect your own assistant

Any MCP client (Claude Desktop, Claude Code, an in-house agent) can connect with an access token.

The token is separate from your invite link; ask for one.