Drawdown anatomy for EA developers: balance vs equity, depth, duration and recovery

Drawdown is the number that switches EAs off

Nobody abandons a trading bot because its profit factor dipped from 1.6 to 1.4. They abandon it at the bottom of a drawdown, usually a week before it recovers. For a developer, drawdown is the statistic that decides whether the Expert Advisor gets to keep running long enough for its edge to show. So it deserves more than one number, and the strategy tester's single "maximal drawdown" line is not enough.

Dissect it into four: which curve you measured, how deep it went, how long it lasted, and how long it took to get back. Then add the one number the tester can never give you: the worst it has ever been, on record, where you cannot erase it.

Balance drawdown vs equity drawdown

Balance moves only when a trade closes. Equity moves every tick, because it includes the floating profit or loss of open positions. A forex robot that closes losers promptly has nearly identical balance and equity curves. An EA that holds losers, adds to them, or runs a grid has a smooth balance staircase sitting on top of an equity curve with holes in it.

Worked example: balance peaks at $10,000. The EA opens a basket that goes $3,200 underwater before recovering and closing for +$150. Balance drawdown from this episode: zero. Equity drawdown: 32%. A reader who only sees balance thinks the EA is flawless. The broker, who was holding margin against that $3,200, knows better.

ConfirmedTrades computes drawdown on the time-weighted growth index so deposits cannot mask a fall, folds the live equity stream into that index, and keeps both figures: Max DD (balance) and Max DD (equity), in currency and percent, in the Glance hub. If the two differ by much, the strategy holds losers. Maximum drawdown explained covers the reader's view; as the developer you should know the gap before anyone else does.

Depth: the percentage, and what it costs to climb back

Depth is peak to trough as a share of the peak. The arithmetic of getting back is not symmetric and new developers underestimate it: the gain needed to recover is the drawdown divided by (1 minus the drawdown). A 10% fall needs 11.1% to recover. 20% needs 25%. 33% needs 50%. 50% needs 100%. The drawdown recovery calculator does this for you, free.

The Score's risk component puts 15 of its 25 points on maximum drawdown, scaling from full marks at 0% to nothing at 50%, and the AI review adds a watch flag above 40%. Those thresholds are not arbitrary; past 50% the recovery gain exceeds the whole account and most strategies never make it.

Duration and recovery: the Ulcer index and recovery trades

Two EAs with a 20% maximum drawdown are not equally painful. One spent three days there; the other spent five months. Depth alone cannot tell them apart, which is why the Performance depth card shows two more numbers. Ulcer index is the root mean square of the drawdown curve measured at every trade, so it rises with both depth and time spent underwater. Lower is calmer. A strategy with many shallow short dips can have a lower Ulcer index than one with a single long 15% trough.

Average recovery and worst recovery count the trades taken to climb from the start of a drawdown back to a fresh high, across every completed drawdown spell in the record. Suppose there are 14 completed spells, average recovery 22 trades, worst 131 trades. If the EA takes 8 trades a week, the worst spell was four months. That is the figure to put in the product description, because it is the one a buyer will actually live through. The card also lists the five deepest spells.

Ratios that put drawdown next to return

• Recovery factor: net profit divided by max drawdown in currency. $3,000 net on a $1,000 worst drawdown is 3.0. The Score gives full credit for this part at 5.0.
• Calmar: annualised return divided by max drawdown percentage. A 40% year on a 20% drawdown is 2.0. Records under a year are not extrapolated, so a three-month record's Calmar uses the raw return.
• Sortino per trade: mean trade return over downside deviation only. It rewards an EA whose losses are small and regular even if its winners are lumpy.
• Tail ratio: best day (95th percentile) divided by the magnitude of the worst day (5th percentile). Above 1 means good days outsize bad days.

The drawdown ledger: the number you cannot reset

In the strategy tester you can rerun until the drawdown looks nice. On a track-record site the old trick was to delete the account after a bad spell, reconnect it, and start the history after the damage. ConfirmedTrades keeps a drawdown ledger keyed to the broker server and login rather than to the platform record. Every sync updates the worst balance and equity drawdown ever seen. Peaks only rise, worst values only grow, and reconnecting the same login picks the ledger straight back up.

For a developer this is a commitment device. The figure in Glance is the real worst case since the account was first connected, and potential buyers are told so. If you change the EA's risk settings, start a fresh broker account for the new version rather than hoping the old drawdown fades; it will not.

Designing for the next drawdown, not the last one

The historical maximum is a floor. The future contains market conditions the history does not, and the Monte Carlo in the Risk Lab makes that concrete by reshuffling your own trades 1,000 times and reporting the 95th-percentile drawdown alongside the median. A common planning rule is to assume the next maximum drawdown will be at least 1.5 times the historical one, and to choose lot sizes so that you would keep the EA running through it. Sizing is its own subject: position sizing and risk of ruin for algo traders is next.