Two Algos, Same Annual Return, Completely Different to Trade: Drawdown Depth, Losing Streak Length, and Time Under Water
Two strategies show up in your backtest folder. Both returned about 34% on a $50,000 account last year. On a spreadsheet, they're twins. Fund them both in an evaluation account and one of them will get you paid while the other one gets you shut down — and it has nothing to do with the return.
Annual return is the number every strategy vendor leads with, and it's the least useful number for deciding whether an algo is tradeable in a prop firm account. What actually decides that is the shape of the ride: how deep the equity curve digs, how many losses arrive back to back, and how long the account sits below its old high-water mark. Three numbers. Here's how to read them before you arm anything.
Number one: drawdown depth versus your firm's actual limit
Max drawdown is the largest peak-to-trough drop in the equity curve. Two 34% algos can post very different ones — say 8% for the first and 22% for the second. That 22% strategy isn't "riskier in theory." On a $50,000 evaluation with a $2,500 trailing drawdown, a 22% equity dip is a $11,000 hole. You are out of the account long before the strategy gets its chance to recover.
The comparison you actually need is drawdown depth in dollars against your firm's drawdown rule in dollars, sized to the contract count you plan to run. Do it in that order. And remember that the historical max drawdown is a floor, not a ceiling — the worst drawdown in your backtest is simply the worst one that happened to occur in that sample. Live trading is entitled to produce a bigger one.
A practical rule: if the backtested max drawdown consumes more than about half your firm's allowed drawdown at your intended size, the strategy isn't too aggressive — your size is. We walked through how contract count changes that math in micros or minis for your eval.
Firm rules differ enough that the same algo can be fine at one shop and unusable at another — trailing versus end-of-day drawdown especially. That comparison lives in how each firm's drawdown and payout rules change which bots belong in your lineup.
Number two: losing streak length, the number that breaks traders
Drawdown depth is a dollar problem. Consecutive losses are a psychological one, and they're what actually causes people to disable a working bot.
A strategy with a 62% win rate that takes two trades a day will still hand you a run of six, seven, eight losers — not as a malfunction, but as ordinary math. If you've never looked up that number, the streak arrives feeling like evidence that something broke. You start checking the logs mid-session. You flatten a position manually "just this once." You turn the strategy off on Wednesday and it takes the week's biggest winner on Thursday without you.
So pull the max consecutive losses out of your backtest before you go live, and write it on a sticky note. Then answer the question honestly: when I hit that streak, what am I going to do? The correct answer is "nothing, because I already knew it was coming." That answer is only available to people who looked the number up in advance.
Two algos with identical returns can differ enormously here. A high-win-rate mean-reversion strategy might cap out at four straight losses; a breakout strategy taking small, frequent losses while it waits for one big trend day can string together twelve. The second one is not worse. It is just far harder to sit through, and knowing that ahead of time is what makes it sittable.
Number three: time under water
This is the one almost nobody checks, and it's the one that ends the most funded runs.
Time under water is how long the account stays below its previous equity high before making a new one. Depth answers "how bad did it get." Time under water answers "how long did I have to live there." An algo can post a modest 6% drawdown and then spend four months grinding sideways before it recovers — no disaster, no rule violation, just seventeen weeks of paying platform fees and watching an account that isn't going anywhere.
Two 34% algos might show a longest underwater period of three weeks and five months respectively. The three-week version is fundable. The five-month version, run alone, means you may pass an evaluation and then wait most of a year for a payout that clears the profit target. Payout mechanics compound this — see minimum trading days, request timing, and the rules that get payouts denied.
Reading the three together
These numbers interact, and the combination tells you more than any of them alone:
- Shallow drawdown, short streaks, short time under water. The easy one. Usually a lower-return, higher-frequency strategy. Boring and fundable.
- Deep drawdown, short time under water. Violent but fast — a sharp dip and a sharp recovery. Survivable only if the dip fits inside your firm's limit at your size.
- Shallow drawdown, long time under water. The morale killer. Never dangerous, never productive. Fine as one piece of a portfolio, punishing as your only bot.
- Deep drawdown, long streaks, long time under water. Whatever the annual return says, this is not an evaluation-account strategy.
Where this actually pays off: combining bots
The reason to measure these three separately is that it tells you which strategies pair well. A bot that goes underwater in choppy, rangebound markets and one that goes underwater in fast trending markets will smooth each other's equity curve — their bad months don't overlap. Two bots with the same drawdown shape just give you the same drawdown, twice as big.
That's the whole argument in breakout vs. snapback, and it's why you have to check whether your bots are secretly the same trade, covered in correlation is the hidden risk in your bot portfolio.
Before you arm anything
Pull four numbers from every strategy you're considering, and pull them at the contract size you actually intend to trade:
- Max drawdown in dollars — compared to your firm's drawdown limit in dollars
- Max consecutive losses — written down where you'll see it during the streak
- Longest time under water in calendar days
- Return, last, as a tiebreaker between strategies that already passed the first three
Then confirm it on your own machine rather than the vendor's spreadsheet. Your first two weeks with a bot covers the sim-to-live sequence, and the NinjaTrader 8 Control Center, tab by tab shows you where to watch it once it's running.
Two algos, same annual return. One of them you can actually trade. The difference was never in the return — it was in the three numbers underneath it.
Ready to run strategies that were built with drawdown, streaks, and recovery time in mind? See how Push Button Trading's automated NinjaTrader strategies work.
