Crypto Ulcer Index measures the root mean square of percentage drawdowns across a defined equity sample. Squaring each observation gives deeper losses more weight, while repeated observations below a peak capture persistence. Compare values only when the equity basis, peak convention, observation frequency and measurement window match; the index is neither a maximum-loss limit nor a recovery-time forecast.
Two strategies can both finish where they started and both suffer a 20% maximum drawdown, yet spend very different amounts of the measured sample underwater. Ulcer Index helps describe that difference. Its useful question is: how much squared drawdown was present across the observations we actually measured?
Define exactly what goes into the square
Use a positive, consistently marked equity series that includes open positions and applicable costs. For this article, the reference peak is the highest equity observed from the beginning of the stated sample through the current observation. Equality with the peak gives zero drawdown.
For observation t:
H[t] = maximum of E[0] through E[t]D[t] = 100 × (1 − E[t] / H[t])Ulcer Index = square root of (sum of D[t]² / N)
Here E[t] is equity, D[t] is positive drawdown in percentage points and N includes every scheduled observation, including zero drawdowns. This is a root mean square, not a sample standard deviation: the denominator is N, not N − 1.
Peter Martin’s original Ulcer Index explanation describes squared percentage drawdowns and an algorithm that updates the highest value as the series advances. Its historical investment examples are not crypto performance evidence.
Some software returns drawdowns as negative fractions. Squaring removes the sign, but it does not fix units: a fractional result of 0.07071 corresponds to 7.071%, rather than 0.07071%. The PerformanceAnalytics UlcerIndex documentation gives the drawdown-based root-mean-square definition and explicitly notes sensitivity to observation frequency. Check a package’s output scale before comparing it with a percentage display.
Same worst loss, different observed severity
The following original hypothetical example uses eight daily UTC observations, from Day 0 through Day 7. Both portfolios start with 100 equity units and end with 100. There are no external cash flows; assume all relevant costs and open positions are already reflected in these invented marks. This is a calculation fixture, not a KLINEO result or recommended strategy.
| Observation | Equity A | Drawdown A | A squared | Equity B | Drawdown B | B squared |
|---|---|---|---|---|---|---|
| Day 0 | 100 | 0% | 0 | 100 | 0% | 0 |
| Day 1 | 100 | 0% | 0 | 95 | 5% | 25 |
| Day 2 | 80 | 20% | 400 | 90 | 10% | 100 |
| Day 3 | 100 | 0% | 0 | 80 | 20% | 400 |
| Day 4 | 100 | 0% | 0 | 85 | 15% | 225 |
| Day 5 | 100 | 0% | 0 | 90 | 10% | 100 |
| Day 6 | 100 | 0% | 0 | 95 | 5% | 25 |
| Day 7 | 100 | 0% | 0 | 100 | 0% | 0 |
| Sum | — | — | 400 | — | — | 875 |
Both series have a 20% maximum observed drawdown and a 0% endpoint return. A has one below-peak observation; B has six.
For A, UI = sqrt(400 / 8) = 7.071%, rounded. For B, UI = sqrt(875 / 8) = 10.458%, rounded. The larger figure describes B’s greater average squared drawdown in this particular sample. It does not mean B will lose 10.458% tomorrow.
The squaring matters. One 20% drawdown contributes 400 to the sum, the same contribution as four 10% drawdowns. Averaging and taking the square root then restores the original percentage scale.
For the single deepest loss and recovery denominator, use the maximum-drawdown guide. For exact elapsed peak-to-recovery time, use the DCA drawdown-duration worksheet. Ulcer Index compresses a series; it cannot reconstruct those episode timestamps.
Sampling can erase the event
Now retain only Day 0 and Day 7 from portfolio A. Both observations are 100, so both drawdowns are zero and the calculated Ulcer Index becomes 0%. The 20% loss on Day 2 has disappeared from the dataset, despite remaining part of the hypothetical underlying path.
The example shows why a weekly endpoint report can miss a short crypto drawdown. It does not establish that coarse sampling always lowers the index: changing the sampled peaks and the denominator can alter the result in different directions.
Do not convert a daily Ulcer Index into an annual value by multiplying by the square root of 365. The inputs are losses relative to peaks, not independent one-day return deviations being scaled through time. Recalculate on the intended observation series and label the result.
A fair comparison preserves the calendar window, timestamp zone, interval and valuation method. Equal-frequency observations make persistence contribute through repeated marks. An irregular event-only series instead gives more weight to periods with more recorded events. A time-weighted variant would be a separate declared calculation, not automatically the formula above.
Missing marks are especially consequential. Forward-filling a stale price can make a holding look flat through a dislocation; dropping missing rows can change the weighting. Preserve the gap, state the policy and use the backtest data-quality audit before accepting a smooth result.
A rolling high is a different reference
A chart indicator may compare each close with a trailing high and then average squared drawdowns over a rolling period. StockCharts documents a 14-period closing-price implementation. That is one platform convention, not a universal crypto setting.
Suppose equity falls from 100 to 80 and stays at 80. With the full-sample running peak defined here, each subsequent observation remains 20% underwater. In a trailing-high implementation, once the old 100 observation leaves the high’s window, the reference can become 80. New drawdown observations then become zero even though the original capital has not recovered. The rolling average can subsequently decline as older squared losses leave its own window.
Record both windows if the high and averaging periods differ, as well as the warm-up rule. Also state whether a pre-window peak is carried into a reporting slice. Resetting the reference at the first equity mark in every slice can hide an already ongoing drawdown.
Use a comparison card before choosing a strategy
| Field to retain | Required evidence or decision |
|---|---|
| Measured object | Portfolio equity, strategy equity or asset price; these are different series |
| Capital changes | Unitized performance series or another declared flow adjustment |
| Peak rule | Running peak from a stated origin, carried-in peak or trailing high |
| Observation schedule | Exact interval, timezone, missing-mark policy and observation count |
| Report window | Fixed start and end, including warm-up and initial observation rules |
| Companion evidence | Return, maximum depth, episode timestamps and current open exposure |
Deposits should not manufacture recovery. Build the appropriate equity series using the cash-flow-adjusted drawdown ledger before applying this formula.
Completed-trade statistics describe another part of the evidence. Profit factor and payoff ratio summarize winners and losers without preserving this equity path. The Calmar ratio guide examines return relative to maximum drawdown; its denominator answers a different question from average squared drawdown.
Appending zero-drawdown observations can lower Ulcer Index without changing the historical worst loss. A zero result on a flat or steadily rising observed series therefore establishes only that those marks show no drawdown under the chosen reference. An empty sample is undefined, and a zero reference peak makes the percentage formula undefined. If equity reaches zero or becomes negative, flag the capital-loss condition explicitly rather than treating the resulting number as a routine risk comparison.
Method and limits
This AI-assisted article combines linked primary definitions and implementation documentation with an original eight-observation example. The sums, square roots and endpoint-sampling comparison were recomputed. No customer history, live trading system or KLINEO Ulcer Index feature was assessed. A lower historical value cannot establish future safety, liquidity or a suitable loss allowance.