3. How much drawdown is normal in a strategy that works

Risk 3: Drawdown and ruin

Now comes the uncomfortable moment in this level. The question is not how to avoid drawdown, because you cannot. The question is how much drawdown is normal, and the answer surprises almost everybody: quite a lot more than you were prepared to accept.

The historical bar

Before judging a strategy it is worth looking at what it is being compared with. These are falls in whole markets, diversified by definition, with no leverage and with nobody making bad decisions:

• The United States, 1929-1932. The Dow Jones lost around 89% from its peak in September 1929 to the low of July 1932. It did not recover that nominal level until 1954: some twenty-five years.
• The United States, 1973-1974. The S&P 500 fell close to 48%, and in real terms the damage was worse because of the inflation of those years.
• The technology index, 2000-2002. The Nasdaq lost around 78% in two and a half years and took some fifteen years to get back to its nominal peak.
• The financial crisis, 2007-2009. The S&P 500 fell around 57% in seventeen months and set a new high in 2013, a little over four years after the low.
• The pandemic, 2020. A fall of close to 34% in barely twenty-three sessions, a little over a calendar month, recovered in some five months.

There is no broken strategy anywhere in that list. It is what a stock index does when you give it enough time. Any system built on equities inherits that behavior, and often amplifies it.

The figure you ought to expect

A reasonable working rule: a strategy with an expected return of 8% or 10% a year and a volatility of 15% or 20% will see falls of 30% or 40% with some regularity, and once a generation something worse. Antti Ilmanen puts it more elegantly in his work on risk premiums: expected returns are not a gift, they are the payment for putting up with exactly these episodes. If nobody suffered, there would be no premium to collect.

Two consequences worth accepting all at once. The first is that a 30% or 40% fall is not evidence that your strategy has broken; it is evidence that your strategy is a strategy. The second is that the worst drawdown in your record is not a ceiling, it is simply the worst that came up in the sample you happened to look at. The prudent course is to plan with a considerably larger figure, of the order of one and a half times it, because the future holds more sample than the past.

A record with no suffering in it is a warning sign

Here is the twist almost nobody internalizes. When somebody shows you twenty simulated years at 18% a year with a maximum drawdown of 6%, the right reaction is not calm, it is suspicion.

A twenty-year period of real data contains 2008 and 2020. That a strategy came through them barely ruffled admits few benign explanations and a great many bad ones:

• Overfitting. Hundreds or thousands of parameter combinations were tried and the one that best dodged the specific falls in that sample was published. That skill does not travel into the future.
• Look-ahead bias. The simulator used data the trader did not have at the time, such as accounting earnings published months later but dated to the quarter they refer to.
• Survivorship bias. The test universe contains only companies that still exist today.
• Missing costs. No commissions, no spread, no market impact and execution at the closing price.

López de Prado has spent years pointing out that when you try many configurations on the same data, finding a spectacular one stops being a discovery and becomes a statistical certainty. Smoothness is not a virtue of the method: it is usually the fingerprint of having hunted for it too hard.

Time under water hurts more than depth does

A great deal is said about how far you fall and very little about how long it takes. Time under water is the number of months or years you spend below your previous peak, and in practice it is what decides who quits.

Compare two paths. The first falls 45% and recovers its peak in fourteen months: it is brutal, but it is lived through as an accident with an ending. The second falls 22% and takes six years to come back: no particular day ever hurts much, and it is six years looking at a statement worse than the one from a while back, six years explaining it to your partner or your clients, six years asking yourself whether this still works or simply stopped working. The second destroys far more portfolios than the first, and in the return tables it looks like the good one.

The historical record confirms it without ambiguity: what hurt about 1929 was not the three years of falling, it was the twenty-two years of waiting that followed. What hurt about the Nasdaq was not 2001, it was 2009 with the index still at half its peak of 2000.

Real case: half the capital, twice

Berkshire Hathaway has one of the best-documented long records in existence. Its share price fell close to half between mid-1998 and March 2000, while the market was rewarding the technology companies the firm had refused to buy, and it fell roughly by half again between late 2007 and early 2009.

An investor who had judged Warren Buffett by his drawdown in March 2000 would have concluded that he was finished and had arrived late to the new economy. That conclusion, held on the basis of perfectly correct data, would have cost twenty-five years of subsequent returns. That a strategy loses half its value tells you almost nothing about whether it is any good. What it does tell you is everything you need to know about what you have to be prepared to put up with.