Drawdown Analysis

In one sentence

Drawdown analysis measures the drops in your bank from its highest point so far, telling you how deep and how long the bad spells are likely to be, even for a winning strategy.

How it works

Profit and ROI tell you where you end up. Drawdowns tell you what the journey feels like. A strategy that makes £800 over a season but spends months £500 below its best is much harder to stick with than the headline suggests.

A drawdown starts when the bank falls from a peak and ends when it climbs back above that peak. The two numbers that matter are the maximum depth, and the time to recover. Losing streaks are the everyday version of the same thing.

The key idea is to work out, before you start, how big a drawdown is normal for your edge and odds. Then when one arrives you can tell bad luck from a broken model.

The maths

• Bank with subscript t is your bank after bet t.
• Peak with subscript t is the highest bank reached up to bet t.
• DD with subscript t is the current drawdown as a share of that peak; for flat stakes it is often quoted in £ or points instead.
• MaxDD is the worst drawdown over the whole period.

In plain English: drawdown is how far below your best you are right now, and max drawdown is the worst that ever got.

Worked betting example

A simple example first. Your bank peaks at £1,450 and falls to £1,160. The drawdown is (1,450 − 1,160) ÷ 1,450 = 20%, or £290.

Now what is normal? Say you back home wins on Match Odds at 3.00 with a true win rate of 36% (an illustrative edge), £10 flat stakes, before commission. Each bet has an expected profit of 0.36 × £20 − 0.64 × £10 = £0.80.

1. Expected profit over 1,000 bets: £800.
2. At £0.80 per bet, recovering the £290 drawdown above takes about 363 bets on average.

We simulated 20,000 runs of 1,000 bets with this edge:

So a winning strategy with an 8% edge will typically give back £300 at some point, one run in ten gives back almost £500, and a losing run of 13 in a row is normal. Despite the edge, 3.9% of runs were still behind after 1,000 bets.

The losing run figure matches a quick rule: the longest run of losses in n bets is roughly log(n × p) ÷ log(1 ÷ q), which gives 13.2 here.

Where it's good

• Setting realistic expectations before you start a strategy.
• Sizing your bank so that a normal drawdown does not force you to stop.
• Telling bad luck from a broken model: a drawdown well beyond the simulated 95th percentile is a warning sign.
• Comparing staking plans: fractional Kelly and smaller flat stakes shrink drawdowns.
• Judging tipsters and trading bots on more than their headline profit.

Limitations and pitfalls

• The simulated ranges assume you know your true edge. If the edge is smaller than you think, real drawdowns will be deeper and longer.
• A backtest shows only one historical path, which is usually luckier than average because the strategy was chosen for looking good.
• Longer odds mean much deeper drawdowns and longer losing runs for the same edge.
• Max drawdown grows the longer you bet, so compare figures only over the same number of bets.
• Correlated bets, such as several matches in the same Saturday 3pm round all backed on one model view, deepen drawdowns beyond what an independent-bet simulation shows.
• Percentage drawdowns and £ drawdowns tell different stories under percentage staking; state which one you mean.

How to build it

• In pandas: running peak with cummax, drawdown as peak minus bank, then take the maximum.
• Simulate thousands of paths in numpy using your real odds distribution, edge estimate and commission to get the normal range.
• Record the date and length of every drawdown, not just the worst.
• Practical tip: write down, before you start, the drawdown at which you will stop and review the model, and stick to it.

Related methods

• Risk of ruin – the extreme drawdown that ends the bank.
• Fractional Kelly – a staking choice that reduces drawdowns.
• Monte Carlo simulation – how to get the normal range.
• Sharpe and Sortino ratios – risk-adjusted returns that complement drawdown.
• Value at risk – a short-horizon measure of bad days.