Backtest metrics explained: CAGR, volatility, Sharpe ratio, and max drawdown

A backtest tear sheet reports how a portfolio would have behaved over a historical period, and it does so with a handful of summary numbers. Four of them, compound annual growth rate, volatility, Sharpe ratio, and maximum drawdown, appear on nearly every tear sheet in the industry. This guide defines each one, shows how it is computed, gives a sense of what a reasonable value looks like for a diversified portfolio, and explains how each misleads when read alone.

The card below is a live tear sheet of a 60/40 portfolio on ENSEMBLE. Its four figures are the ones defined on this page, and you can open the full model to see every metric in context.

Compound annual growth rate (CAGR)

Definition. CAGR is the single yearly rate of return that would turn the portfolio's starting value into its ending value over the backtest period, with compounding. If a portfolio grew from 100 to 250 over 12 years, its CAGR is 250 divided by 100 raised to the power 1/12, minus one, which is about 7.9 percent a year.

Why it matters. CAGR is the headline. It answers the question most people ask first: how fast did the money grow?

How it misleads. CAGR contains no information about the path. A portfolio that compounds at 8 percent with a 15 percent worst drawdown and one that compounds at 8 percent with a 50 percent worst drawdown have the same CAGR and are not the same investment. CAGR is also extremely sensitive to the start and end dates: the same strategy measured from 2007 and from 2009 can differ by several percentage points a year, entirely because of what happened in 2008.

Reasonable range. Over a twenty-year window that includes 2008, US large-cap equities have compounded at roughly 8 to 10 percent, a 60/40 portfolio at roughly 6 to 8 percent, and Treasury bills at 1 to 2 percent. A strategy claiming a CAGR far above equities over that window with equity-like risk deserves suspicion.

Never compare a CAGR from one period to a CAGR from another. Compare the strategy to a benchmark measured over the same dates, which is what the tear sheet's benchmark line is for.

Annualized volatility

Definition. Volatility is the standard deviation of the portfolio's daily returns, scaled to an annual figure by multiplying by the square root of the number of bars in a year: 252 for an exchange calendar, 365 for a market that trades every day. A daily standard deviation of 1 percent corresponds to annualized volatility of about 15.9 percent on an exchange calendar. ENSEMBLE reads the calendar from the series, so a crypto strategy is not understated by the 252 convention.

Why it matters. Volatility measures how bumpy the ride was. Two portfolios with the same CAGR and different volatility took different amounts of risk to get there, and the lower-volatility one used its risk more efficiently.

How it misleads. Volatility is symmetric: a sharp gain raises it exactly as much as a sharp loss. A strategy that has frequent large up days and small down days will show high volatility even though an investor would have been happy with every one of those days. Volatility also describes typical dispersion and says nothing about the tails. A portfolio can have modest volatility for years and then lose a third of its value in a month.

Reasonable range. US equities have run at roughly 15 to 20 percent annualized volatility over the last twenty years, with spikes above 40 percent in 2008 and 2020. A 60/40 portfolio typically shows 9 to 12 percent. Long Treasuries alone run about 13 to 15 percent, which surprises people who think of bonds as low risk.

Sharpe ratio

Definition. The Sharpe ratio is the portfolio's annualized return in excess of the risk-free rate, divided by its annualized volatility. If a portfolio returned 8 percent while Treasury bills returned 2 percent and its volatility was 12 percent, its Sharpe ratio is (8 minus 2) divided by 12, or 0.5. William Sharpe introduced the measure in 1966 as the reward-to-variability ratio.

Why it matters. Sharpe puts return and risk on one scale, so portfolios with different risk levels can be compared. It is the standard measure of risk-adjusted return and the one most often used to rank strategies.

How it misleads. Because it uses volatility as the risk measure, the Sharpe ratio inherits all of volatility's problems. It penalizes upside variance, ignores tail risk, and is flattered by strategies that earn small steady returns and occasionally lose a lot. It also depends on the risk-free rate used and on the window; a strategy's Sharpe ratio over a period when T-bills yielded 5 percent is not comparable to one over a period when they yielded zero.

Reasonable range. Over long windows, a Sharpe ratio between 0.4 and 0.8 is typical for a long-only, unlevered, diversified portfolio. The S&P 500's long-run Sharpe ratio is roughly 0.4 to 0.5. Sustained Sharpe ratios above 1.0 are rare for liquid, unlevered strategies over twenty years and usually indicate either a favorable sample or overfitting. Sharpe ratios above 2.0 in a backtest are almost always a mistake.

Maximum drawdown

Definition. Maximum drawdown is the largest percentage fall from a peak in the portfolio's value to the lowest point before a new peak, over the whole backtest. If the portfolio reached 200, fell to 120, and later recovered, the drawdown for that episode was 40 percent. The maximum drawdown is the worst such episode.

Why it matters. Drawdown is what an investor actually experiences. Nobody feels volatility; everyone feels watching a portfolio fall by a third. It is the metric most closely tied to whether a strategy would have been abandoned at the worst moment, which is the most common way real investors underperform their own backtests.

How it misleads. Maximum drawdown is a single episode and depends entirely on whether the sample contains a severe bear market. A backtest from 2009 to 2019 will show a modest maximum drawdown for almost any equity strategy, not because the strategy is safe but because the period was. It also says nothing about how long the drawdown lasted. A 30 percent decline recovered in six months and a 30 percent decline that took five years to recover have the same maximum drawdown and feel entirely different.

Reasonable range. Over a window including 2008, US equities drew down about 55 percent. A 60/40 portfolio with long Treasuries drew down roughly 20 to 30 percent, depending on the bond sleeve and the exact dates. Trend-following strategies such as the 200-day moving average rule have historically shown maximum drawdowns in the 15 to 25 percent range on US equities, which is their main appeal.

Look at the full drawdown chart, not just the maximum. It shows every episode, how long each lasted, and whether the strategy's worst periods coincided with the benchmark's.

Turnover and trade count

Definition. Turnover is the fraction of the portfolio's value traded over a period, usually a year. The trade count is the raw number of buy and sell transactions. ENSEMBLE reports both, along with rebalances per year in the model's run metadata.

Why it matters. Turnover is where costs live. Every trade pays a spread and possibly a commission, and in a taxable account every sale can trigger a gain. A strategy that trades daily needs a far larger gross edge than one that trades quarterly to come out ahead after costs. Turnover is also a check on whether the rule was implemented as intended: a strategy described as "rebalanced monthly" that shows six trades a year is not rebalancing monthly, or is only trading when weights change.

Reasonable range. An annually rebalanced fixed allocation has turnover well under 20 percent. Monthly rebalancing of the same allocation trades every month but in small amounts, so turnover stays low while trade count is high. A monthly momentum rotation like dual momentum has turnover of 100 to 300 percent a year. Anything above that is a trading strategy, not a portfolio.

Time in market

Definition. Time in market is the share of the backtest period during which the strategy held risk assets rather than cash.

Why it matters. It separates tactical strategies from buy-and-hold. A trend-following rule that spent 30 percent of the period in Treasury bills earned nothing from equities during that time, and its return has to be understood in that light. A high CAGR with low time in market means the strategy earned its return in fewer months, which is impressive and also means the result depends heavily on getting those months right.

Start date, end date, and benchmark

These are not metrics, but every metric above is meaningless without them. The start date is where the tear sheet begins: by default the first day two assets in the universe have data, so a rule that chooses between assets is never judged while it can only hold one, or the date you asked for in the description. The end date is the most recent session; the benchmark is the passive alternative the strategy is being compared to. A tear sheet that omits any of the three cannot be compared to anything, and comparing metrics across tear sheets with different dates is one of the most common backtesting mistakes.

Reading them together

No single metric describes a strategy. A workable order is:

1. Drawdown first. Could an investor have held this through its worst period?
2. CAGR against the benchmark over the same dates. Did the strategy earn its complexity?
3. Sharpe ratio. Did it earn its return efficiently?
4. Turnover. What did it cost to run, and does that cost survive taxes?
5. Time in market and holdings over time. What did the rule actually do?

If the strategy still looks good after those five, the next step is to change one thing about it and see whether the result is robust. The how to backtest a portfolio guide covers that loop. The prompt below builds the 60/40 model on the card above; change the weights, the bond fund, or the rebalancing schedule and watch which metrics move.

Backtests are hypothetical, past performance does not guarantee future results, and ENSEMBLE is research software rather than an investment adviser.