Reference

Analytics Glossary

What does this number on my screen actually mean — and what should I do differently knowing it? That is the only question this page answers. Every term below is defined in plain business language, with no formulas and no statistics degree required, and each definition ends where it should: at the decision it ought to change. It comes in two halves. Platform analytics covers the vocabulary shared across the forecast, accuracy, risk, hedging and positioning screens. Scenario metrics covers the measures specific to War Room runs. Every term has its own anchor, so any page — or any colleague — can link straight to a definition rather than to the top of this one.

Platform analytics

The vocabulary that recurs across the forecast, accuracy, risk, hedging and positioning screens.

Backtest

A backtest is how a forecast earns your trust before you spend money on it. The platform takes a forecast frozen at a point in the past and compares it against the prices that actually arrived afterwards — genuine out-of-sample validation, not a flattering fit to data the model had already seen. It is what the Forecast Scorecard's track record is built from. Check it before you lean on any forward number, and check it for your commodity rather than for the platform as a whole: see Forecast Accuracy & Why You Can Trust the Data.

MAPE, RMSE and MAE

These three grade how close past forecasts landed, and lower is better in all of them. MAPE is the headline: how far the forecast sat from the real price, on average, as a percentage, counting misses in both directions — it reads like a business KPI, which is why it leads. MAE and RMSE say the same thing in price units, per barrel or per tonne, and RMSE punishes large misses harder. That difference is the useful part: an RMSE sitting well above its MAE tells you the commodity occasionally throws a big surprise rather than drifting steadily off, which is a reason to hold cushion even when the average error looks comfortable.

There is no universal threshold for a good MAPE, and any table claiming one would mislead you. What counts as accurate in a calm, well-supplied market would be remarkable in a thinly traded one that can swing sharply inside a single month, so the honest way to use these figures is comparative: this commodity against its own record, and this horizon against the shorter one, rather than against a fixed band. The Forecast Scorecard is where that comparison is published, commodity by commodity — look up the one you buy at the horizon you actually plan on in Forecast Accuracy & Why You Can Trust the Data.

Confidence band

The shaded range drawn around a forecast line. Instead of a single number it says: we are about 90% confident the price lands somewhere inside this range, and 90% is the level drawn throughout the platform — so the upper edge you budget against has a known meaning. The band is narrow in calm markets and fans out as volatility rises and as the horizon lengthens, which is the platform being honest about accumulating uncertainty rather than a defect. Plan against the band, not the centre line: for a buyer the upper edge is the prudent worst-case cost, and for a seller the lower edge is the conservative revenue assumption.

Calibration coverage

A quality check on the bands themselves: a 90% band should contain the real price about 90% of the time, and calibration coverage measures whether it actually has. Coverage sits on the Calibration tab of the Forecast Scorecard, and how to read it is set out in Forecast Accuracy & Why You Can Trust the Data.

Market regimes

A regime is the market's personality right now: how jumpy prices are, and which way they are trending. Every commodity is classified into one of four states, and the state tells you how much weight to put on a smooth forecast line and how fast you need to move. A calm regime lets you plan on a deliberate timeline; a turbulent one argues for acting sooner and holding more cushion than the forecast alone would suggest. Regimes are set in the wider risk picture, alongside the measures that size your downside, in Market Regimes & Risk Analytics.

HOLD and DE-RISK

The action cue shown beside the regime. HOLD means conditions are stable and no change of posture is needed. DE-RISK means the market has moved into a riskier state: a caution flag to review exposure and consider trimming, hedging or buying forward. It is not an automatic instruction to sell. The practical reading of a fresh DE-RISK is one of timing — it is a reason to pull next week's decision into this week.

Value-at-Risk and Expected Shortfall

These two size your downside so you can hold the right cushion against it. Value-at-Risk (VaR) is the worst move you would expect at a stated confidence level; the platform shows 95% and 99% versions. As a worked example — hypothetical figures, not a performance claim — a 95% VaR of 5.2% means that in a bad month, the worst 1-in-20, this price could move about 5.2% against you. Expected Shortfall, also labelled CVaR, answers the follow-up: once that line is breached, how ugly does it typically get? It is always at least as large as VaR, which makes it the more conservative number to plan against. Buyers turn these into budget headroom; traders turn them into position size.

GARCH-VaR and VIX-adjusted VaR

Two dynamic flavours of the same measure, both labelled on the Risk Dashboard. GARCH-VaR reacts to recent shocks — tight in calm markets, widening quickly after a spike — so it reflects today's conditions rather than a long-run average. VIX-adjusted VaR re-scales the estimate to broad market stress, the market's fear gauge, so it widens when everything is nervous and not only when your commodity is. The signal worth acting on is the gap: when either dynamic figure sits well above the plain baseline, conditions have changed, and you should size off the dynamic number rather than the average.

StatArb

Short for statistical arbitrage: a market-neutral strategy that trades two commodities whose prices are statistically tethered to each other. When the gap between them stretches unusually wide you buy the cheap leg and sell the expensive one, betting the gap closes. Because you hold both sides, the profit comes from the gap reverting rather than from the market going up or down — which also means the real risk is that the tether itself breaks. The three entries that follow are how you judge whether it will hold.

Cointegration and hedge ratio

Cointegration identifies two prices that move together over the long run: they drift apart briefly but reliably snap back. It is a stronger claim than plain correlation, which can be a coincidence that evaporates, and that durability is what makes it worth committing money to. The hedge ratio is the dosage — how many units of commodity B to trade against each unit of A so the two largely offset, where 0.8 means roughly 0.8 units of B per unit of A. “Hedge beta” on the StatArb screen is the same idea under a different label. Get the ratio wrong and you are not hedged, you are simply holding two positions.

Z-score and half-life

Both describe the gap between two tethered commodities. The z-score says how stretched that gap is against its own normal range: near zero is ordinary, and the further the gap stretches from zero the more unusual it is — the platform flags the rows stretched far enough to be worth a look, which is the classic mean-reversion cue. Half-life says how many days the gap has typically taken to close half the distance back to normal. Read them together, because a stretched gap with a long half-life is a position you will be carrying for months: match the half-life to the horizon you can actually hold before you act on the z-score.

COT positioning

COT is the weekly Commitments of Traders report, which splits open futures positions into commercials — producers and consumers hedging real-world exposure, the so-called smart money — and speculators, who are betting on direction. Watching the balance shows you who is leaning which way, and the most actionable read is a crowded one: heavily one-sided speculative positioning flags a market vulnerable to a sharp reversal if the news turns. Treat that as a reason to size smaller and set a tighter exit, not as a reason to abandon the idea.

Contango and backwardation

The shape of the futures curve — what delivery later costs against delivery sooner. In contango, later contracts cost more, so rolling a position forward costs you something each time, and it usually signals comfortable supply. In backwardation, later contracts are cheaper, rolling forward earns a small premium, and it usually signals tight near-term supply. Curve shape therefore sets the real cost of holding a position over time, as well as being a live read on how tight the market is. Check it before you put on anything you intend to hold: time is either working for the position or against it.

Brier score

A grade for probability calls — a statement such as “70% chance of an up month”. It sits on the Regimes tab of the Forecast Scorecard, and how it is graded is set out in Forecast Accuracy & Why You Can Trust the Data.

Scenario metrics

These measures appear on War Room run results. Every one of them is a comparison rather than a level: the same simulation run twice over real market data, once with your event and once without, so a scenario number always answers one specific question — what would this event add on top of whatever the market was going to do anyway?

Scenario, shock, intensity and duration

A scenario is a named what-if: one or more shocks plus a horizon, twelve months by default. Scenarios are saved, re-runnable and shared across your organisation. A shock is one discrete event inside it — a conflict, an embargo, a demand slump — carrying a size, a duration and an intensity. Intensity is how fully the event lands: 100% is full force, 50% a half-strength version. Duration is how many months it keeps pressing before fading. Between them they let you dial a textbook crisis down to the version you actually consider plausible, which is the version worth planning against.

Baseline

Every scenario is measured against a baseline: the same market data and the same simulation, run once without your event. Every impact number you read is the difference between those two worlds, which is what lets you attribute an effect to your shock rather than to market movement that would have happened regardless. It is also why re-running the same scenario next month can give a different answer — the baseline has moved, so quote the run date whenever you quote the number.

Baseline μ and Shocked μ

μ, pronounced “mu”, is the screen's shorthand for the typical month-by-month path a price is expected to take. Baseline μ is that path without the shock and Shocked μ is the path with it. The gap between the two is the number to take into a budget conversation, because it is the part of a projected cost change your scenario is genuinely responsible for. Quoting the level instead of the difference is the most common way a scenario gets over-read.

σ Mult

σ, pronounced “sigma”, stands for how much prices swing around their expected path, and σ Mult is the multiplier your shock applies to that swing: 1.5× means half again as much uncertainty as normal. The trap is to check the likely price and stop reading. A scenario can barely move the expected path while sharply raising σ Mult, and a market that has become harder to plan around is itself a reason to hold more cushion or hedge earlier, even when the central forecast looks untouched.

Δ Median

Δ is shorthand for “change”, and Δ Median is the headline number of every run: the most likely price change caused by the shock alone, expressed as the percentage gap between the shocked and baseline most-likely end prices. It is what the impact tables rank by, and it is the figure to quote when someone asks what a scenario does to copper. For procurement the arithmetic is immediate — multiply Δ Median by your annual spend on that commodity and you have a first-cut budget delta to argue from.

Fan chart and confidence intervals

The engine replays the coming months over and over, once as a world without your shock and once with it, then stacks the outcomes into a fan. CI means confidence interval — the same idea as the confidence band on a forecast chart. The dark inner core, labelled Shocked 50% CI, holds the central half of the simulated futures; the pale outer fan, Shocked 95% CI, holds nearly all of them; and the dashed Baseline median sits alongside for comparison. Read the outer edge as your prudent worst case. It is the cushion this scenario would demand, and it is usually more useful to a budget owner than the headline.

Hop

Hop counts how many steps a commodity sits from the original shock as the effect ripples outward along relationships measured from real price history. A hop-1 market is hit through one direct relationship; a hop-2 market is hit by the ripple's ripple, and the effect weakens at every step away from the source. The reason to read the column at all is that a hop-2 impact is usually an exposure nobody had written down — and undocumented exposures are the ones that surprise a budget.

Regime change

A regime is the market's current mood, calm or turbulent and rising or falling; the four states are set out above under market regimes. The Regime column and the dashed halos on the ripple map mark where a scenario tips a market from one mood into another — a calm uptrend into a turbulent decline, say. A flip is often more consequential than the price move that caused it, because it changes how that market behaves for months rather than for a week, and it changes how much you can rely on any forecast of it.

Scenario VaR and CVaR

Scenario-conditional versions of the measures defined above under Value-at-Risk and Expected Shortfall. VaR 95% asks how far the price could move against you across this scenario's worst 1-in-20 simulated outcomes; CVaR 95% asks how deep the move typically goes once you are in that worst 1-in-20, and is always the harsher of the two. Both size the cushion you would need if the event happened, not on an average day. Portfolio VaR and Portfolio CVaR roll the same pair up across every affected commodity while counting each one equally, regardless of how much of it you actually buy — a deliberately coarse whole-book read, good for asking how much riskier the scenario makes everything together, and not usable as an absolute figure for your own positions.

Leader, Stability and the payoff matrix

Producer-group quota scenarios are not taken at face value. They are settled by a simulated negotiation in which one side moves first and the others respond in their own best interest — a strategic game between the largest producers of that commodity, the Leader being whoever moves first. Stability says how likely the modelled agreement is to hold: high means it sticks, low means someone is tempted to cheat, so an announced cut tends to leak and the real supply effect arrives smaller than the press release. The Leader payoff matrix is a what-if grid scoring every combination of moves, and its figures are a comparative ranking of those moves, never a profit forecast. The practical use is narrow and valuable: Stability tells you how much of an announced quota shock to actually plan against.

Hedge ticket

A hedge ticket on a run's Strategy tab is a pre-structured protective trade, sized on the same dosage idea set out above under cointegration and hedge ratio. It names a long leg to buy, a short leg to sell and a Ratio of one against the other, plus an Entry price where the position would be opened, an Exit where you take it off once it has done its job, and a Stop where you abandon it because the market has proved the idea wrong. What a ticket buys you is a conversation: it turns “we are exposed” into something a treasury or a desk can actually assess. It is AI-drafted decision support to validate against your own book and limits, not an instruction, and the run's risk warnings sit directly beneath it for that reason.

Monitoring KPI

A real-world tripwire: the early observable sign that a scenario is starting to happen — a freight rate, an inventory report, a policy announcement. Each run's Strategy tab lists three to five of them. Put them on your weekly watchlist, because they are the difference between a scenario you filed and a scenario you are standing guard over: they are what tell you when to stop treating the plan as hypothetical and start executing the one you rehearsed.

Projection

The Projection banner is the honest fallback. Where a commodity does not carry enough price history for the full simulation, the view degrades to a coarser directional estimate — which way prices lean and roughly how far — and the banner tells you so on the face of the run. Treat those figures as orientation and never as numbers to budget against. Commodities with too little history are left out of a run entirely rather than guessed at, so the banner is the platform drawing the distinction between a weak answer and no answer.