How deep can a portfolio fall? Reading drawdown with real numbers

Every portfolio app shows you how much your portfolio has grown. Almost none shows you the other half of the same history: how far it fell along the way, how long it stayed down, and how often that happened. This guide works through those three questions on one real portfolio — the demo portfolio, four invented companies with six years of generated prices, measured by the same engine as a real account.

The figures below are the portfolio as it actually was on each day, rebuilt from its trades — not today's holdings replayed over the past.

The one number everyone quotes

Between January 2019 and December 2024 the demo portfolio returned 84% in total, about 10.8% a year. Over the same six years its maximum drawdown was −39.4%.

Maximum drawdown is the largest fall from a previous high to a later low. For this portfolio the high came on 17 March 2021 and the low on 21 March 2022. One unit invested at the start had grown to 1.49 at the peak; a year later it was worth 0.90. Everything gained since 2019 had gone, and some of the money put in with it.

• V_t — the portfolio's value on day t
• M_t — the highest value it had reached on any day up to t
• MDD — the most negative of those daily gaps, over the whole period

Two things about this number are worth knowing before leaning on it. It is measured from the peak, not from what you paid, so a portfolio can be comfortably in profit and still be 30% below its best. And it describes one single episode: the worst day of the worst fall. It says nothing about how often falls happen, or how long they last.

A 39% fall needs a 65% rise

The arithmetic of recovering is not symmetrical, and it is the reason drawdown hurts more than its number suggests.

A portfolio that falls 39.4% keeps 60.6% of its value. To get back to where it was it has to rise by 39.4 / 60.6 — that is, 65%. A 20% fall needs a 25% rise; a 50% fall needs a 100% rise. The deeper the fall, the steeper the climb back, because the climb starts from a smaller base.

How long it lasted

Depth is what gets reported. Duration is what makes people sell.

The demo portfolio's deepest fall took 369 days from peak to low, and another 499 days to recover: from March 2021 to August 2023, 868 calendar days below its previous high. Counted in trading sessions, the longest stretch under water was 619 — about two and a half years in which the portfolio was, every single day, worth less than it had been at some earlier point.

That is the figure to hold next to the −39.4%. A fall of that depth that recovers in six months is uncomfortable. The same fall spread over two and a half years is the one that breaks people, because by the time the bottom arrives nobody believes in the recovery any more. The measure is on time under water.

A second number from the same calculation is even more telling: the demo portfolio was below a previous high on 94% of its trading days. It set a new high on roughly one day in seventeen. That is not a sign of a bad portfolio — it is what almost every volatile portfolio looks like from the inside, and it is invisible on a chart of growth.

It was not only one fall

The same history holds five separate falls of more than 12%:

Fall	From its peak on	Down in	Back up in	Total
−39.4%	17 Mar 2021	369 days	499 days	868 days
−24.9%	7 Aug 2023	95 days	178 days	273 days
−20.7%	8 Apr 2020	133 days	89 days	222 days
−15.2%	3 Dec 2024	28 days	still under water	28 days
−12.4%	22 Jan 2019	34 days	116 days	150 days

In six years, three falls deeper than 20%. Reading only the maximum drawdown, you would have seen one.

The typical bad stretch

Because maximum drawdown is a single episode, it is unstable: one different month in 2022 and the number changes a lot. The conditional drawdown answers a steadier question — how bad are the bad stretches, on average?

It takes every day's distance below the previous high, keeps the worst 5% of them, and averages those. For the demo portfolio it is −29.0%. In plain terms: when this portfolio was in one of its bad stretches, it was typically about 29% below its best. That is a number you can plan around; the −39.4% is the tail of it.

The Ulcer index goes one step further and folds depth and duration into one figure, by averaging the square of every day's drawdown. The demo portfolio scores 12.5%. A portfolio with a single sharp fall that recovers quickly scores low; one that spends years 20% down scores high even if it never touches −40%. Two portfolios with the same maximum drawdown can have very different Ulcer indices, and the one with the higher index is the one that was harder to hold.

Putting it next to the return

Drawdown on its own says how much a portfolio can hurt. Next to the return it says whether the hurt was paid for.

The Calmar ratio divides the annual return by the maximum drawdown: 10.8% a year against a 39.4% fall is about 0.27. Each point of annual return came with almost four points of worst-case fall. There is no universal threshold for a good Calmar ratio, but it is the right way to compare two portfolios whose returns look similar: the one that got there with the shallower falls is the one more people would actually have stayed in.

What drawdown does not tell you

• It is backward-looking. A −39.4% maximum drawdown over six years does not cap the next fall at −39.4%. The 2008 crisis took broad equity indices down by more than 50%, deeper than anything in the demo's window.
• It depends on the window. Start the measurement in April 2022, after the worst of it, and the maximum drawdown looks much milder. A drawdown figure without its dates is incomplete.
• It is not a loss unless you sell. The distance to the peak is real, but it only becomes a realised loss if the position is closed at the bottom — which is exactly what long, deep drawdowns make people do.

Doing this with your own portfolio

Everything above comes from the drawdown chart and the Stress tests tab of the dashboard. Import a broker CSV, or type a few trades, and the same figures are computed for your own history — including the table of your own worst falls, with dates. Or open the demo and look at the figures above in place.