Drawdowns and losing runs

The question

"I've lost 12 in a row and I'm down for the month. Has my edge gone, or is this just a bad run?"

That's the most expensive question in betting, and getting it wrong in either direction costs money. This lesson shows what betting drawdowns and losing runs look like for a strategy that genuinely wins, and how to tell a normal bad run apart from an edge that has actually died.

The idea in one sentence

Even a real edge produces long losing runs, deep drawdowns and months without a new high, so you must know before you start what normal bad luck looks like, and judge the edge during a bad run by closing line value, which is far less noisy than profit.

The picture

We simulated 20,000 bettors, each with a genuine +5% edge (before commission) at odds of 3.0, placing 1,000 bets at level stakes of £10. Every one of them is a winning bettor. Here's what their journeys looked like.

Read the last line twice. A typical winning bettor goes more than 400 bets, several months for most people, without their bank reaching a new peak.

The chart draws 20 of those bettors as faint lines, with the median in bold. Some shoot up, some wander below zero for hundreds of bets, and a few are still underwater at the end. All of them have the same real edge.

Now the dead edge

Run the same simulation with the edge set to zero (fair odds, so commission makes it slightly losing). Over 1,000 bets the median worst drawdown is £550, the median longest losing run is 15, and 61% finish in a loss.

Compare that with the live edge: a live bettor's 1-in-10 bad drawdown (£645) is deeper than a dead bettor's typical one. Drawdowns and losing runs overlap heavily between live and dead edges. That's why P&L alone can't tell you which one you've got.

Worked Betfair example

You've been backing at an average of 3.0 in Match Odds and Over/Under markets for 400 bets at £10 level stakes. You believe you have a +5% edge. You're down £300 (30 points). (Illustrative figures; probabilities from 200,000 simulated bettors.)

1. What a live edge expects. After 2% commission the edge is +3.6% per bet. Over 400 bets that's +14.4 points on average, with a standard deviation of √400 × 1.41 ≈ 28.2 points.
2. How unusual is −30 for a live edge? It's 44 points below the average, about 1.6 standard deviations. In simulation, 5.0% of live-edge bettors were at −30 or worse after 400 bets.
3. How usual is it for a dead edge? A dead edge averages −5.3 points over 400 bets. 17.3% of dead-edge bettors were at −30 or worse.
4. What the P&L tells you. The result is about 3.5 times as common under "dead" as under "alive". That tilts the odds, but if you thought the edge was probably real before, you still can't be confident it has gone. The P&L alone is not enough.
5. Check the closing line value. Over those same 400 bets, did you get better prices than the Betfair closing price? CLV is measured on every bet whatever the result, and it swings far less.
   • Say each bet's CLV has a standard deviation of about 10% (illustrative). Over the last 200 bets the standard error is 10% ÷ √200 ≈ 0.71%.
   • Average CLV +4%: that's 5.7 standard errors above zero. You're still consistently beating the market's final view. This is almost certainly a bad run. Carry on at your planned stakes.
   • Average CLV about 0%: you're no longer beating the close. The market has caught up with whatever you were doing. Treat the edge as dead: stop, and go back to research.
6. Check your tripwires. Before starting, you should have written down: "Review if the drawdown passes £740 (the 1-in-20 level for 1,000 bets) or CLV over the last 200 bets drops below +1%." At −£300, neither is hit.

Verdict: a £300 drawdown after 400 bets is uncomfortable and completely normal for a real +5% edge. The P&L can't settle the question; the CLV can, and it can do it within a few hundred bets rather than a few thousand.

The plan, before the first bet

1. Simulate your own numbers. Use your real average odds and a cautious edge in the tool above. Note the typical and the 1-in-20 losing run and drawdown.
2. Size your stakes to survive the 1-in-20. If the 1-in-20 drawdown is 74 points, a bank of 50 stakes isn't enough (Lesson 6.3).
3. Write down your tripwires. A drawdown level for "review", a CLV level for "stop". Decide them now, not while you're losing.
4. Track CLV on every bet. It's your early-warning system (Lesson 2.4).
5. Use a proper change test for the final call. A CUSUM chart flags a real shift in performance without over-reacting to noise (Lesson 7.3).

This is the heart of how Statometrics works: an edge is only worth having if you can tell, honestly and early, the difference between a normal losing run (ride it out) and an edge that has died (kill it without sentiment).

The formula

Typical longest losing run

• n is the number of bets.
• p is your win rate and q = 1 − p your loss rate.
• ln is the natural logarithm.

In plain English: for 1,000 bets at a 35% win rate, ln(350) ÷ ln(1/0.65) ≈ 13.6. Expect a losing run of about 14 somewhere in every 1,000 bets. It grows slowly with n, but it's there even for great bettors.

Chance of still showing a loss after n bets

• μ (mu) is expected profit per £1 staked (0.036 after commission here).
• σ (sigma) is the standard deviation of one £1 bet (1.41 here).
• Φ is the bell-curve area to the left of a point (Lesson 1.1).

In plain English: the edge grows with n but the noise only grows with √n, so the chance of being behind shrinks slowly. After 1,000 bets: Φ(−0.81) ≈ 21% after commission, 13.5% before. The simulation's 20% and 13% agree.

Drawdown

• Bank_s is your bank after bet s.
• max takes the highest bank reached so far.

In plain English: how far you are below your best-ever point. Your worst drawdown is the biggest this ever gets, and it's the number that tests your nerve.

Try it

Set odds 3.0, a +5% edge and 1,000 bets, and check the chance of still being in a loss comes out around 13 to 14% before commission (about 20% after), with a median longest losing run of about 14. Then set the edge to 0% and compare the drawdowns: notice how much the two pictures overlap.

Common mistakes

• Judging the edge by the last month's P&L. A few hundred bets of P&L barely separates a live edge from a dead one. CLV does it far faster.
• Deciding the tripwires during the drawdown. Every rule you make while losing is shaped by fear. Set them before the first bet.
• Stopping at the bottom of a normal run. Quitting a genuine edge after 14 straight losers locks in the loss and throws away the edge. That run was expected.
• Riding a dead edge on hope. The opposite mistake. If CLV has fallen to zero and stayed there, no amount of patience brings the edge back.
• Staking so big a normal drawdown ends the game. If the 1-in-20 drawdown would empty your bank, the stakes are wrong, however good the edge.

The flagship article on why real edges are so hard to hold on to: why good models still lose money.