Part I · Returns and risk · Chapter 1D

Risk-adjusted return

The Sharpe ratio, and what it cannot see

The single most consequential thing to take in before you evaluate any short volatility strategy is this: the Sharpe ratio cannot see skew. A strategy can win 95% of the time and still lose money, as Chapter 1C showed. A strategy can show a Sharpe ratio of 0.8 in a backtest and a Sharpe ratio of 0.2 in a live retail account. The gap between those two numbers is, mostly, the shape of the return distribution: specifically the third and fourth moments, which the Sharpe ratio ignores by mathematical construction. If you remember only one thing from Part I, remember that.

Chapters 1B, 1C and this one make a single argument in seven steps:

1. A strategy can win often and still lose money (Chapter 1C).
2. This happens when small gains are frequent but rare losses are large (Chapter 1B).
3. That means mean, variance, skewness and kurtosis all matter (Chapter 1B).
4. The Sharpe ratio sees only the mean and the variance.
5. Short options are structurally short skew.
6. Therefore short volatility strategies can look better than they are.
7. Therefore every short volatility trade needs an analysis of its tail loss and of its margin path.

This chapter develops steps 4 to 7.

1 · The Sharpe ratio

A strategy that earns 10% a year with 5% annual volatility is different from a strategy that earns 10% a year with 30% volatility, and the Sharpe ratio formalizes the difference. It is the ratio of expected excess return to the standard deviation of returns. It answers the question of how many units of expected return a strategy pays you for each unit of risk it asks you to take:

The numerator is the excess return: what you earn above the risk-free rate $r_f$, the return on cash lent to the government. The denominator $\sigma_R$ is the standard deviation of the returns. By construction the ratio is dimensionless and scale-invariant. Doubling your position size doubles both the excess return and the volatility, which leaves the Sharpe ratio unchanged. That is what makes it useful for comparing strategies of different sizes.

With a hypothetical risk-free rate of 2%, the two strategies above have Sharpe ratios of $(0.10 - 0.02)/0.05 = 1.6$ and $(0.10 - 0.02)/0.30 = 0.27$. Their expected returns are the same, and their Sharpe ratios differ by a factor of six.

For short-horizon intraday strategies, the kind most of the modeling chapters of Part III assume, the risk-free return over one bar is negligible, so the practical form drops it:

Annualizing a Sharpe ratio

Chapter 1A showed that over $n$ independent periods the mean return grows $n$ times and the standard deviation grows $\sqrt{n}$ times. Their ratio therefore grows $n/\sqrt{n} = \sqrt{n}$ times, so a daily Sharpe ratio converts to an annual one with the square root of 252 trading days:

A strategy whose daily returns average 0.05% with a daily standard deviation of 1.26% has a daily Sharpe ratio of $0.05/1.26 = 0.040$, and an annual one of $0.040 \times 15.87 = 0.63$. This is the famous rule that the Sharpe ratio scales with the square root of frequency, and it is the structural reason that high-frequency strategies report eye-popping Sharpe ratios (Chapter 3D).

2 · What the Sharpe ratio adds to expected value

Two strategies can have identical expected returns but very different Sharpe ratios if their volatilities differ. The lower-volatility strategy is preferable for three concrete reasons.

It compounds faster. From the variance drag of Chapter 1A, the compound growth rate is about $\mu - \sigma^2/2$. With the same arithmetic mean, the strategy with the lower variance compounds to more terminal wealth: 10% with 5% volatility compounds at $0.10 - 0.00125 = 9.9\%$, and 10% with 30% volatility at $0.10 - 0.045 = 5.5\%$. Variance drag is real money over multi-year horizons.

It can carry leverage. A strategy with a Sharpe ratio of 0.5 cannot safely be leveraged four times. The leveraged version still has a Sharpe ratio of 0.5 but four times the volatility, so its routine drawdowns become catastrophic. A strategy with a Sharpe ratio of 4.0, typical of intraday market making, can be, because its volatility was small to start with and its drift was overwhelming. The maximum sustainable leverage is, roughly, a function of how confident you are that the next move is positive, which is what the Sharpe ratio measures. The section after next makes this exact.

It is easier to live with. Picture two strategies that each win 51% of their trades. One has tight bounds and a high Sharpe ratio, and it feels safe to operate. The other has wide bounds and a low Sharpe ratio, and it feels terrifying. It has long losing streaks, you cannot tell in real time whether it is working or just lucky, and you are likely to abandon it during a drawdown. The path of a high-Sharpe strategy is psychologically easier to ride than the path of a low-Sharpe strategy with the same expected return.

The figure below simulates one year of daily returns for four strategies, all with an expected annual return of 8% and all driven by the same random shocks. Their daily volatilities are scaled so that their Sharpe ratios are 0.5, 1.5, 3 and 6, which means annual volatilities of 16%, 5.3%, 2.7% and 1.3%. The only control is a seed that re-draws the shocks, starting at 7. The readouts give each strategy's return over the year.

seed

#7

SR 0.5, terminal = 3.6%

SR 1.5, terminal = 7.0%

SR 3, terminal = 7.7%

SR 6, terminal = 8.0%

At seed 7 the four strategies end the year at +3.6%, +7.0%, +7.7% and +8.0%. The Sharpe-6 curve follows its 8% drift almost in a straight line, while the Sharpe-0.5 curve swings widely around it. Now move the seed to 20, and the Sharpe-0.5 strategy ends the year down 17.0% and the Sharpe-1.5 strategy down 0.5%, although both still expected +8%. A year like that is exactly when a trader abandons a low-Sharpe strategy.

3 · Trading frequency and the Sharpe ratio

Here is one of the most important and least understood facts in quantitative trading. If you have a tiny edge per trade, say a 51% win rate at a fixed size, you do not want to trade it once. You want to trade it many times, because the Sharpe ratio of a strategy grows as the square root of the number of independent bets:

where $\text{SR}_1$ is the Sharpe ratio of one bet and $\text{SR}_N$ that of $N$ independent bets added together. The reason is the one behind (1D.3): $N$ bets add $N$ times the mean but only $\sqrt{N}$ times the standard deviation.

Trading frequency turns a tiny edge per trade into a high Sharpe ratio, which then lets leverage and compounding do their work. This is the structural reason that a high-frequency strategy in the style of Renaissance Technologies can produce double-digit annual Sharpe ratios on a win rate of 51% to 52%. It is not that each bet is highly profitable. It is that each bet is slightly profitable and there are tens of thousands of bets a day. The same edge, traded once a year, would be statistically indistinguishable from luck, and certainly unsafe to leverage.

The retail implication is sobering, because most retail strategies trade only one to ten times a month. Take a bet that wins $1 51% of the time and loses $1 otherwise: its mean is $0.02 and its standard deviation is almost exactly $1, so its per-trade Sharpe ratio is 0.02. Traded 10 times a month, 120 times a year, it has an annual Sharpe ratio of $0.02 \times \sqrt{120} = 0.22$. Even if each trade has a real edge, which most do not, the resulting Sharpe ratio is modest, leverage is constrained and compounding is slow. There is a fundamental advantage to high frequency that retail traders cannot match. Trying to overcome it by adding leverage to a low-frequency strategy is exactly the path to ruin that Chapters 13A and 13C warn against.

4 · Why a low Sharpe ratio caps leverage

Suppose your strategy has an arithmetic mean return $\mu$ and volatility $\sigma$ a year, and you lever it by a factor $k$, holding $k$ dollars of it for each dollar of capital. Ignoring the cost of borrowing, the levered strategy has:

• an expected return of $k\mu$;
• a volatility of $k\sigma$;
• a Sharpe ratio still equal to the original: leverage does not change it.

What leverage does change is the probability of ruin: the probability that the equity curve ever falls to some fraction $f$ of its starting value. For a strategy whose wealth follows geometric Brownian motion (Chapter 1A), that probability is

when the exponent is positive, and 1 otherwise. Here $\mu$ and $\sigma$ are the drift and volatility of the levered strategy. Levering up multiplies $\mu$ by $k$ but $\sigma^2$ by $k^2$, so the ratio $2\mu/\sigma^2$ is divided by $k$. Higher leverage means a lower exponent, and a lower exponent drives the probability of ruin toward 1.

Where the ruin formula comes from

Step 1: follow the logarithm of wealth. By Chapter 1A, the log of wealth drifts upward at $G = \mu - \sigma^2/2$ a year, with volatility $\sigma$.

Step 2: use a standard result. A quantity that drifts upward at a rate $G > 0$ with volatility $\sigma$, and moves continuously, ever falls a distance $a$ below its starting point with probability $e^{-2Ga/\sigma^2}$. This is a standard property of Brownian motion with drift, which Part VII introduces. If $G \le 0$ it reaches any level below eventually, with probability 1.

Step 3: translate the distance. Wealth reaching a fraction $f$ of its start means the log of wealth falling $a = -\ln f$. Substituting that distance into the probability gives

$$\begin{aligned} e^{2G\ln f/\sigma^2} &= f^{\,2G/\sigma^2} \\ &= f^{\,2\mu/\sigma^2 - 1} \end{aligned}$$ (1D.6)

Step 4: add leverage. With leverage $k$, the exponent becomes $2k\mu/(k^2\sigma^2) - 1 = (2\mu/\sigma^2)/k - 1$. It falls as $k$ rises, and reaches zero when $k\sigma = 2\mu/\sigma$.

The exponent in (1D.5) is positive only while the levered volatility stays below twice the Sharpe ratio. With a risk-free rate of zero, the condition $k\sigma < 2\mu/\sigma$ reads $k\sigma < 2\,\text{SR}$. A strategy with a Sharpe ratio of 0.5 stops growing at all at 100% levered volatility, and grows most quickly at 50%, the Kelly level that Chapter 3D derives. That ceiling assumes normal returns, and the fat tails of Chapter 1B put the practical ceiling lower. The Sharpe ratio sets the leverage ceiling. No amount of clever risk management overrides this.

5 · The Sharpe family

The basic Sharpe ratio has several variants worth knowing. Each changes what counts as risk.

The Sortino ratio

The Sortino ratio uses only downside volatility in the denominator:

Here $\sigma_{\text{down}}$, the downside deviation, is computed only from returns below some threshold $\tau$, often zero or the risk-free rate:

Every return above the threshold counts as zero, so only shortfalls add to the risk. The Sortino ratio correctly recognizes that upside volatility is not "risk": nobody complains about big gains.

Because it leaves out the upside, the Sortino ratio is usually larger than the Sharpe ratio. For a normal distribution with a 5% mean and 10% volatility, the downside deviation below zero is 4.6%, so the Sortino ratio is 1.09 against a Sharpe ratio of 0.5. The gap depends on where a strategy's volatility sits. The more of it is on the upside, the more the Sortino ratio exceeds the Sharpe ratio. The CBOE PUT Index has a Sortino ratio approximately equal to its Sharpe ratio, because most of its volatility is downside volatility: there is not much upside surprise in collecting a fixed premium. SPX has a Sortino ratio noticeably higher than its Sharpe ratio, because some of its volatility comes from upside surges.

The Sortino ratio fixes one defect of the Sharpe ratio: it stops penalizing upside surprises as risk. It captures one part of the skew problem, but it still does not capture kurtosis, or the depth of the left tail, explicitly.

The information ratio

The information ratio measures excess return against a benchmark instead of the risk-free rate:

where $R_s$ is the strategy's return and $R_b$ the benchmark's. The denominator is the volatility of the tracking error, the variation of the strategy's returns around the benchmark's. The information ratio is the right measure for active portfolio managers who are judged against a benchmark and care about how much active risk they take to produce active return.

The Treynor ratio

The Treynor ratio divides excess return by beta instead of by the standard deviation:

Beta is the sensitivity of the strategy's return to the market's, $\beta = \operatorname{Cov}(R_s, R_m)/\operatorname{Var}(R_m)$, where $R_m$ is the market's return. The Treynor ratio is used when you assume that the risk specific to one position can be diversified away and only market risk matters. It is less used than the others, because it borrows its assumptions from the capital asset pricing model (CAPM), the textbook model in which market risk is the only risk that is paid. Those assumptions do not hold well empirically.

The Calmar ratio

The Calmar ratio divides the mean annual return by the maximum drawdown, the largest peak-to-trough loss in the record, which Chapter 1E defines formally:

It penalizes path-dependent risk directly. It asks: per unit of pain you had to absorb to take part in this strategy, how much were you paid? Rankings by the Calmar ratio often differ sharply from rankings by the Sharpe ratio.

The Calmar ratio also catches what the Sharpe ratio misses entirely. XIV, a short volatility exchange-traded product terminated in 2018, had a Sharpe ratio of about 1.5. Its Calmar ratio is hardly meaningful: a maximum drawdown of about 96% destroyed the strategy. For strategies focused on preserving capital, the Calmar ratio is the most honest single-number measure, and the ranking it gives tells a more honest story than the Sharpe ratio does.

Drawdowns and the cost of recovery

Drawdown is one-sided and real, and it binds on whether a portfolio survives. Chapter 1E derives the gain needed to recover from a drawdown $D$, which is $D/(1 - D)$. A 40% drawdown needs a 67% gain to break even, a 50% drawdown needs 100% and a 60% drawdown needs 150%. The cost keeps accelerating: a 70% drawdown needs 233%, an 80% drawdown 400% and a 90% drawdown 900%.

Recovery cost is non-linear in the size of the loss, which is why discipline about maximum drawdown binds on survival. A strategy with a positive expected return but a 60% drawdown, one that would have happened anyway in your backtest, needs 150% from the trough to get back to even. At the strategy's normal compound rate, that recovery can take a decade or more. Meanwhile, the opportunity cost of capital sitting in a deep drawdown is the return you could have earned in a simpler, lower-drawdown strategy over those same years. Survive small losses and you can compound. Suffer a single large one and you may never compound back to your prior peak within a useful career.

This is also why the academic literature distinguishes drawdown-adjusted measures of performance from volatility-adjusted ones. One strategy produces its volatility through small, frequent moves around a rising mean, which gives a high Sharpe ratio. Another produces the same volatility through occasional catastrophic losses with smooth gains in between, the short volatility shape. The two are fundamentally different. The Sharpe ratio reads them the same. The Calmar ratio, the Sortino ratio, the expected shortfall and the maximum drawdown all read them very differently.

Expected shortfall and tail beta

Chapter 1B introduced the expected shortfall, $\text{ES}_{95}$: the average loss in the worst 5% of periods. For sizing a short volatility position, it is the single most useful number you can compute.

Tail beta is the strategy's beta to the market given that the market is in its worst 5% of months:

Here "tail" means the months with $R_m \le \text{VaR}_{95}$ for the market. For most premium-selling strategies, tail beta is approximately 1.0. Your "diversifier" or "income overlay" then moves one for one with everything else in your portfolio precisely when you needed diversification most. The CBOE PUT Index has a tail beta near 1.0 to SPX in the worst months, and so does most of the "alternatives" allocation of a typical endowment portfolio. This is the empirical content of the saying that all correlations go to one in a crisis, which Chapter 1F examines.

The PUT Index in numbers

Professor Oleg Bondarenko of the University of Illinois Chicago analyzed the index in a Cboe-commissioned white paper. He covered the 32 years from June 30, 1986 to December 31, 2018, and Bondarenko, 2019 gives these headline numbers:

The skewness is several multiples worse than that of SPX itself. The PUT Index also delivered a Sortino ratio higher than those of SPX, the MSCI World index, the Russell 2000 and 30-year Treasury bonds over the same window.

Through 2024 the index continued to deliver, with a 2024 calendar-year return of 17.8% against 25.0% for the SPX Total Return index. Over the 10 years to September 2025, it delivered 8.0% a year at lower volatility than the index. The strategy works. But understanding why it works, that it is paid for accepting negative skew and not for any clever timing, is the prerequisite for sizing it correctly.

6 · Same Sharpe ratio, different risk

The Sharpe ratio uses the first moment, $\mu$, in its numerator and the square root of the second, $\sigma$, in its denominator. The third and fourth moments appear nowhere, so two strategies with identical Sharpe ratios can have wildly different risk. Chapter 1B built two such strategies from monthly returns and compared their tails. Here is the same point in annual terms, looking at the worst year and the road back from it.

7 · What the Sharpe ratio cannot see

Look again at (1D.1), which contains the mean (the first moment), the standard deviation (the square root of the second) and a risk-free rate. It does not contain skew, kurtosis, the shape of the left tail, autocorrelation, the depth of drawdowns, the time to recover or the dependence on market regime. The Sharpe ratio is a two-parameter summary of a distribution that needs four parameters to describe even roughly, and many more to describe honestly.

This would not matter if real return distributions were approximately normal (Gaussian). They are not. The strategies that retail traders care about include short premium, trend following, factor tilts, volatility-targeted equity and momentum. For most of them, the higher moments are big enough that ignoring them amounts to financial malpractice.

Short options are structurally short skew

As Chapter 1B showed, a short option collects a premium, which caps the right tail, in exchange for a potentially larger loss in an open left tail. The payoff is structurally asymmetric, so the distribution is structurally negatively skewed. This is true for:

• naked short puts;
• short calls;
• short iron condors, which are short volatility on both sides;
• short strangles;
• any premium-selling strategy.

It is also true, to lesser degrees, for:

• defined-risk spreads, whose skew is bounded but still very negative;
• calendar spreads, in which the short front-month option is the dominant Greek;
• most "income" strategies.

Short volatility means short skew. There is no way around this.

So the Sharpe ratio over-rates short volatility

Put the steps together: short volatility has a negatively skewed payoff, and the Sharpe ratio ignores skew. Therefore the Sharpe ratio systematically over-rates short volatility. The bias is consistent: short volatility strategies keep showing attractive Sharpe ratios even when they are a bad allocation.

This is not an indictment of harvesting the variance risk premium as a strategy, and that premium is real (Chapter 10C). The point is that the measure you use to evaluate a strategy has to match the shape of that strategy's distribution. The Sharpe ratio does not match the shape of short volatility. The Calmar and Sortino ratios, and explicit analysis of paths and stress, match it better.

If you allocate by Sharpe ratio alone, you over-allocate to short volatility relative to a properly risk-adjusted portfolio. The fix is not "do not trade short volatility". It is "use better measures".

Most hedge fund alpha is short skew

The key paper is by Jakub Jurek, of Princeton and the National Bureau of Economic Research (NBER), and Erik Stafford, of Harvard Business School. It should be required reading for anyone judging a managed-futures or hedge fund pitch. It appeared in the Journal of Finance as Jurek and Stafford, 2015, *Journal of Finance* 70(5), 2185–2226. They examined broad hedge fund index returns from 1996 to 2012 and asked a deceptively simple question. How much of the apparent alpha, the return beyond what standard risk factors explain, is real? And how much is compensation for accepting negative skew that traditional factor models do not see?

Their finding, restated in its cleanest form: the hedge fund return series is not reliably distinguishable from the returns of mechanical S&P 500 put-writing strategies. Price the skewed payoffs correctly, using a stochastic discount factor that recognizes downside risk in a way traditional linear models cannot, and the apparent alpha largely disappears. A stochastic discount factor is a pricing rule that weights each outcome by how painful it is to lose money in that state. A loss in a crash then counts for more than the same loss in a calm year. The alpha in question is the 6% to 10% a year, before fees, that hedge fund indices appear to earn over 1996 to 2012.

Read bluntly, the result says that once you account for the price of accepting negative skew, the hedge fund industry in aggregate is selling insurance, not generating alpha. The implication is uncomfortable. Institutional sophistication brings global research teams, prime brokerage relationships, custom risk systems, decades of pedigree and $20 billion under management. Suppose all of that cannot produce a skew-adjusted Sharpe ratio meaningfully above what mechanical at-the-money put writing on SPX delivers. Then a retail trader's prior belief that they can do better than put writing, through clever entry timing, smart strike selection or proprietary indicators, should be very, very low.

Put honestly, a retail trader running cash-secured at-the-money XSP puts with mechanical monthly rolls earns roughly the same after-tax expected risk-adjusted return as a sophisticated multi-billion-dollar hedge fund. That holds if both are sized appropriately. Most of what separates the institutional return from the retail one is fees, not skill.

Bakshi, Crosby and Gao (2024, Mathematical Finance) extended the Jurek-Stafford analysis into a formal decision framework for an investor who is averse to ambiguity and to downside risk. They show that the apparent attractiveness of naive short volatility strategies largely disappears once realistic aversion to crashes is priced in. Even if the variance risk premium is real, and it is, a sufficiently risk-averse investor would not raise their expected utility by harvesting it without significant tail hedging or sizing discipline.

The size of the skew tax

How much should you discount a short volatility Sharpe ratio? Several approaches give a range:

• Compare the PUT Index's skewness of −2.09 with SPX's of −0.81. The put-writer is roughly 2.5 times more negatively skewed, which calls for a meaningful discount.
• Apply a Sharpe penalty adjusted for risk aversion: roughly a 30% to 50% haircut.
• Compare Calmar ratios, which capture some of the path. The PUT Index's 0.29 against SPX's 0.18 is a 60% premium, not the 30% premium its Sharpe ratio shows.

The operational rule of thumb follows: mentally apply a 30% to 50% haircut to the Sharpe ratio of any short volatility strategy before comparing it with long-only or trend-following strategies.

8 · How sure are you? The probabilistic Sharpe ratio

A Sharpe ratio computed from a backtest is an estimate from a finite sample. A subtle and important variant asks how confident you can be that the true Sharpe ratio of your strategy exceeds some threshold $\text{SR}^*$, given the sample you have seen. The probabilistic Sharpe ratio (PSR), from David Bailey and Marcos López de Prado, answers it.

First, the standard error of an observed Sharpe ratio $\widehat{\text{SR}}$, measured per period over $n$ observations:

Here $\gamma_3$ is the sample skewness and $\gamma_4$ the sample kurtosis (not the excess kurtosis) of Chapter 1B. Then the PSR is the probability, under a normal approximation, that the true Sharpe ratio exceeds $\text{SR}^*$:

where $N(\cdot)$ is the standard normal distribution function, the probability that a standard normal variable falls below its argument. For normal returns, $\gamma_3 = 0$ and $\gamma_4 = 3$, and the numerator under the square root becomes $1 + \widehat{\text{SR}}^2/2$. Note that $\widehat{\text{SR}}$ here is per period, not annualized: use the monthly Sharpe ratio with monthly data.

The PSR explicitly corrects the standard error of the observed Sharpe ratio for skewness and kurtosis, the moments that the Sharpe ratio itself ignores. Strategies with negative skew, such as premium selling, have wider confidence intervals than strategies with the same observed Sharpe ratio and symmetric returns.

This connects everything in Part I to a single statistical reality. Not only does the Sharpe ratio under-report tail risk, but the estimate of the Sharpe ratio is less reliable when there is more tail risk. Both errors push in the same direction, both flatter premium-selling strategies, and both inflate a retail trader's confidence. Be humble about Sharpe ratios estimated from short backtests of skewed strategies.

9 · The deflated Sharpe ratio: when you tried many strategies

Bailey and López de Prado also introduced the deflated Sharpe ratio, in Bailey and López de Prado, 2014, to address a separate defect. Backtested Sharpe ratios are biased upward, and the bias grows with the number of strategies tested. If you backtest 1,000 strategies, the best one has a phenomenal Sharpe ratio purely from selection. The deflated Sharpe ratio corrects for this. It computes the probability that the observed Sharpe ratio is statistically distinguishable from zero given three things: the number of trials run, the non-normality of the returns and the length of the backtest.

The key idea is the best result that luck alone would produce. Suppose you test $K$ strategies that all have a true Sharpe ratio of zero. Their observed Sharpe ratios scatter around zero, and the largest of $K$ such draws grows with $K$. The deflated Sharpe ratio uses that expected maximum as the threshold $\text{SR}^*$ in the PSR, in place of zero. Your strategy has to beat not zero but the best that $K$ useless strategies would have shown.

The threshold the deflated Sharpe ratio uses

Step 1: the spread of luck. Let $V$ be the variance of the observed Sharpe ratios across the $K$ trials. If every trial's true Sharpe ratio is zero, each observed one is roughly a normal draw with mean zero and standard deviation $\sqrt{V}$.

Step 2: the expected best draw. Bailey and López de Prado approximate the expected maximum of $K$ such draws with the inverse of the normal distribution function, $N^{-1}$, and Euler's constant $\gamma_E = 0.5772$:

$$\begin{aligned} \text{SR}_0 = \sqrt{V}\,\Big[ &(1 - \gamma_E)\, N^{-1}\!\left(1 - \tfrac{1}{K}\right) \\ &+ \gamma_E\, N^{-1}\!\left(1 - \tfrac{1}{Ke}\right) \Big] \end{aligned}$$ (1D.15)

At $K = 100$ the bracket is about 2.53: luck alone is expected to produce a best Sharpe ratio about 2.5 spreads above zero.

Step 3: deflate. The deflated Sharpe ratio is the PSR of (1D.14) with $\text{SR}^* = \text{SR}_0$, and with the standard error of (1D.13), which carries the skewness, the kurtosis and the length of the backtest.

The figure below computes it for a backtest of daily returns, with five sliders:

• The observed Sharpe ratio, annualized, from 0.5 to 5, starting at 1.8.
• The length of the backtest, from 50 to 5,000 daily returns, starting at 252, one year.
• The number of trials run, from 1 to 5,000, starting at 100.
• The skewness of the returns, from −2 to 2, starting at −0.5.
• The kurtosis of the returns, from 3 to 15, starting at 5.

The curve is the probability that the strategy is real at each observed Sharpe ratio, with a dot at the chosen value. A vertical line marks the best Sharpe ratio expected by chance, and horizontal lines mark the 50% and 95% thresholds. Move the trials slider down to 1, then up through 5, 10 and 20.

observed Sharpe

1.80

daily returns N

252

trials run

100

return skew

−0.5

return kurtosis

5.0

max SR by chance = 2.536

z-score vs chance = −0.710

P(strategy is real) = 23.9%

Below threshold. Indistinguishable from a selection-bias artifact.

With 100 strategy variants tried, the best Sharpe expected by chance alone is 2.54. An observed Sharpe of 1.80 counts as evidence of edge only if it beats that benchmark by enough standard deviations to clear the 95% threshold.

At the starting settings, one year of daily returns and 100 trials put the best Sharpe ratio expected by chance at 2.54, well above the observed 1.8, and the probability that the strategy is real reads 23.9%. With 1 trial there is no selection to correct for, and the same 1.8 reads 95.9%, just over the 95% threshold. With 5 trials the chance benchmark is 1.19 and the probability 72.0%. With 10 trials the benchmark is 1.58 and the probability 58.5%, and by 20 trials it is 1.90 and 46.0%. Every added trial slides the whole curve to the right.

Now lengthen the backtest. At 100 trials, five years of daily returns (1,260) lower the benchmark to 1.13 and lift the probability to 92.6%, and ten years (2,520) lift it to 99.9%. Luck's best result shrinks as the sample grows, because each estimate scatters less. One year of data cannot tell a Sharpe ratio of 1.8 from the best of 100 useless strategies; ten years can.

The figure follows the method in the box above, for daily returns. It divides the annual Sharpe ratio by $\sqrt{252}$, takes $V = 1/(N-1)$, the scatter of a zero-Sharpe strategy's estimate over $N$ days, and annualizes the benchmark again for display.

Bailey and López de Prado's empirical finding is sobering: most published "anomaly" Sharpe ratios in the academic literature shrink to indistinguishable from zero after deflation. The retail equivalent is even worse. A trader who has spent six months testing parameter combinations on TradingView, and finally finds one with a Sharpe ratio of 1.2 over a three-year backtest, has run thousands of implicit trials. The deflated Sharpe ratio is almost certainly not significant. Most "discovered" trading systems are statistical artifacts.

10 · Every short volatility trade needs a stress test

The argument ends in an operational rule. Before sizing any short volatility position, compute its $\text{ES}_{95}$ and its expected behavior in stress scenarios. Use falls in SPX of 10%, 20% and 35%, combined with rises in the VIX of 10, 25 and 50 volatility points. This book calls the three combinations S1 (SPX down 10% with the VIX up 10 points), S2 (down 20% and up 25) and S3 (down 35% and up 50), and Chapter 13G builds the portfolio's stress table on them. The Sharpe ratio is, at best, a third-tier input.

The tools for evaluating a short volatility strategy are these:

1. Path simulation. Run the strategy through 2008, 2018, 2020 and 2022 explicitly. What was its maximum drawdown, and how long did it take to recover?
2. Stress profit and loss. What does it lose in S1, S2 and S3?
3. Margin expansion. Under portfolio margin, what cushion do you need to survive S2?
4. Conditional drawdown. What is the average drawdown beyond the 95th percentile, the conditional drawdown at risk of Chapter 1E?
5. Tail beta. What is the strategy's beta to equities in the worst 5% of equity months?

Size any short volatility allocation on these, not on its Sharpe ratio.

The five-question stress test

Before you allocate to any short volatility strategy, answer five questions:

1. What is my tail loss? Specifically, the $\text{ES}_{95}$: the average loss in the worst 5% of months.
2. How does margin expand under stress? Run the position through S2 and see how the broker's margin requirement changes.
3. Can I hold to expiry? If the position marks to a loss of some percentage in 30 days, would I be forced to close it?
4. Is the premium adequate compensation for the skew? Apply the 30% to 50% Sharpe haircut and see whether the strategy is still attractive.
5. What is my kill criterion? It must be specific, measurable and committed to in advance.

Three of these bind hardest. What is the average loss in the worst 5% of months? What maximum drawdown is likely under stress? Can the portfolio recover from that drawdown without behavioral capitulation, the point at which you give up and sell at the bottom? If you cannot answer these in dollar terms for your current portfolio, the position is too big or you do not understand it well enough. Probably both.

11 · Defined risk does not fix the Sharpe trap

Chapter 1B showed that a single spread caps its loss but keeps its negative skew. The iron condor, which sells a put spread and a call spread together, is the case that matters most in practice. It caps your loss per trade, but you take the same negatively skewed bet hundreds of times. The aggregate skew of a many-trade portfolio of condors is still very negative, because trades on the same index are not independent in a crash. Condors also have terrible tail behavior, because both sides of the structure can be threatened in a single move.

The evidence is stark. Cboe's CNDR index, a benchmark for systematic iron condors, returned approximately 690% from 1986 to 2024, against approximately 5,500% for the SPX Total Return index. Over roughly four decades, that is a compound rate of about 5.5% a year against about 11%. It is empirical evidence that systematic condors underperform massively over time. The defined-risk feature does not fix the structural skew.

The two traditions this book draws on disagree here, and both positions stand. Chapter 12C presents the iron condor as an intermediate strategy, with its own uses and mechanics. The evidence in this chapter, and the case in Chapter 10I, argue against running condors systematically in a retail account. Read the three chapters together before you trade one.

12 · The equity risk premium: the benchmark you compete with

Everything in this chapter about Sharpe ratios, tail risk and the chain of retail haircuts sits inside one numerical reality. Over the longest data sets available, equities have paid a real equity risk premium of roughly 4% to 6% a year over the risk-free rate, with extreme variation across countries and centuries. Chapter 10B treats the premium in depth. The canonical studies are these:

• Mehra and Prescott (1985), "The Equity Premium: A Puzzle", Journal of Monetary Economics. Using American data from 1889 to 1978, they documented an average real return on equities of about 7% against about 1% on short-term debt. That 6 percentage-point premium was more than traditional consumption-based asset pricing models could explain without implausibly high risk aversion. The "puzzle" framing is decades old, and the empirical magnitude has held up remarkably well in later data.
• Dimson, Marsh and Staunton, Triumph of the Optimists (2002), and the continuing Credit Suisse and UBS Global Investment Returns Yearbook. They extended the data to 16 or more countries over 100 or more years. Their headline figure for the global equity risk premium outside the United States, relative to bills, is approximately 4.4% a year (a geometric mean, from 1900 onward), with substantial dispersion. Some markets ran higher, among them the United States, Australia and South Africa. Others were materially lower, among them Italy, Japan and Belgium. In some countries, 50-year equity returns were negative in real terms.
• Fama and French (2002), "The Equity Premium", Journal of Finance. They argued that the expected equity premium, derived from dividend and earnings yields, is closer to 3% to 4%. That is meaningfully below the realized historical premium, because much of the realized premium came from unexpected rises in valuations, through expanding price multiples and falling discount rates, that cannot repeat indefinitely.

Three operational points follow.

1. The premium is real but smaller than retail intuition assumes. "Stocks return 10% a year" is a rounded nominal pre-tax figure, not the real after-tax return, net of fees, that you actually compound. The honest expected real after-tax figure for a retail buy-and-hold in American equities is roughly 4% to 5%: meaningfully positive, not a windfall.
2. The choice of countries matters enormously. The most-cited equity risk premium figures, which use American data only, embed a survivorship bias: the United States happened to be the winning market of the twentieth century. The global panel of Dimson, Marsh and Staunton includes markets that lost world wars, suffered hyperinflation, were closed by political revolution or stagnated for decades. It gives a more honest baseline for the premium an investor in 1900 could have expected in advance.
3. The competitor for your active strategy is not zero; it is the equity risk premium. Any active strategy you run has to beat buy-and-hold of a broad equity index after taxes and fees, not beat zero. The four-layer retail haircut of Chapter 13A and the Jurek-Stafford finding point to the same operational conclusion. Once frictions are honestly counted, most retail active strategies fail to beat the structural reward already available from passive equity exposure.

The equity risk premium is the baseline against which everything in this book competes. When a strategy claims edge, the right comparison is not with doing nothing. It is with the combination of the variance risk premium and the equity risk premium that you could harvest passively. That comparison frames the rest of the book.

13 · Where Part I goes next

The lesson of this chapter carries straight into the models of Part III. Building a model with a tiny positive edge per prediction is only the first step. The edge becomes meaningful when it is executed many times, so that the Sharpe ratio rises and leverage becomes usable, and that is the unit of analysis from Part III onward. Part II establishes what you can trade: the products of the SPX, SPY and options complex. It also shows why the choice of product, especially the tax treatment of XSP against SPY (Chapter 2E), is the most consequential operational decision an American retail trader makes.

Chapter 1B gave you distributions, and Chapter 1G explains why edges exist at all. This chapter has shown why the Sharpe ratio misleads for short volatility. Chapter 13A turns all of this into realistic retail expectations with its haircut framework, and Chapter 13B builds the conviction framework that uses it. Without this chapter, the book's judgments of short volatility strategies would seem arbitrary. Why is a Sharpe ratio of 0.65 for the PUT Index not equivalent to a Sharpe ratio of 0.5 for SPX? Now you know.

14 · Check yourself