Maximum drawdown: the number almost nobody publishes
Drawdown is the peak-to-trough fall in your bankroll. It describes what following a strategy actually feels like, decides how large a bank you need, and is the metric prediction services are most likely to omit. How to compute it and how to size against it.
Maximum drawdown is the biggest fall from a high-water mark to a subsequent low. If a bankroll climbs to +14 units, falls to −4, then recovers, the maximum drawdown is 18 units — regardless of where it finishes. It is the most useful single number for deciding how much money a strategy requires, and the most commonly omitted.
How do you compute maximum drawdown?
Walk the equity curve forward, tracking the running maximum, and record the largest gap between that maximum and the current value:
peak(t) = max(equity[0..t])
drawdown(t) = peak(t) − equity(t)
max drawdown = max over all t of drawdown(t)Two conventions, and services rarely say which they use. Absolute drawdown is in units and is comparable across strategies at a fixed stake. Relative drawdown is a percentage of the bank at the peak and is what matters for ruin. Ask which one you are being shown.
Why does drawdown matter more than profit?
Profit is what a spreadsheet experiences. Drawdown is what a person experiences. A service that finishes a season up ten units after being down thirty in March has lost almost everyone who joined in February — and their experience of the identical strategy was uniformly negative.
Two failure modes follow from not knowing the number in advance. Undercapitalisation: a bank sized to the average result cannot survive the worst stretch, and the strategy is abandoned at the bottom. Silent quitting: a follower who was never told a run of five losses is normal reads it as evidence the service is broken.
This is why our own front page prints drawdown and the longest losing run at the same visual weight as profit. Followers do not churn on a bad month; they churn on a bad month nobody warned them about. Publishing the number is cheaper than replacing the customer.
Sizing a bankroll against drawdown
The practical rule used by operators who publish drawdown at all:
bankroll ≈ 1.5 to 2 × historical maximum drawdownAt 1 unit flat with a historical max drawdown of 20 units, hold 30–40 units.The multiplier is not conservatism for its own sake. The historical maximum is a sample minimum of the true worst case: you have observed the worst that happened, not the worst that can happen. A longer future sample will, with high probability, produce a deeper trough than anything in the record.
Higher variance strategies need a bigger multiple. A model betting average odds of 1.60 will have shallower drawdowns than one betting 2.60 at the same edge, because the win frequency is higher and losing runs are shorter.
How long a losing run is normal?
Longer than intuition suggests. For independent bets with win probability p, the chance of a run of k consecutive losses somewhere within n bets rises quickly with n:
| Win rate | Bets | A losing run of this length is unremarkable |
|---|---|---|
| 60% | 100 | 5–6 in a row |
| 60% | 500 | 7–8 in a row |
| 50% | 100 | 6–7 in a row |
| 50% | 500 | 9–10 in a row |
| 40% | 500 | 12–13 in a row |
Read the table before you need it. A run that looks like proof a model has stopped working is, at these frequencies, an ordinary event — which is precisely why a run alone is not evidence either way, and why sample size has to be stated with every claim.
The honest arithmetic on our own record
On the current CONSENSUS sample, the worst drawdown is larger than the total profit. We print that on the front page rather than leaving it to be discovered, because a reader who finds it themselves concludes the page is hiding things, and a reader who is handed it concludes the opposite.
It also means something concrete: on this sample, following every signal at one unit flat required a bank able to absorb a trough deeper than everything the strategy had made. That is the realistic shape of a young record, and it is the reason the front page says a starting point rather than a proof.
Check any of this against our record
Every signal CONSENSUS publishes carries the bookmaker odds fixed before the event starts and the settled result afterwards — including the drawdowns and the losing runs. The running total is on the front page and every entry is in the log.