If you evaluate a trend follower using the Sharpe ratio, you will reject the best compounders and select the ones most likely to disappoint. The metric is not just imprecise. It is precisely backwards
Episode 12 quantified the patience cost. The investor now knows what trend following costs and what it buys. This episode provides the toolkit for the next step: evaluating which trend-following programs to allocate to. The toolkit is not the one the industry typically uses.
The standard evaluation kit, built around the Sharpe ratio, alpha, information ratio, and tracking error, was designed for a Gaussian world: one where returns are normally distributed, volatility is symmetric, and the path does not matter. Trend following exists in a different world, one where returns are positively skewed, drawdowns are asymmetric, and the path determines the outcome. Applying Gaussian tools to a non-Gaussian process does not merely produce imprecise answers. It produces inverted ones. The programs that score highest on the standard kit are not necessarily the best compounders. In some cases they are the worst.
The Sharpe Problem
The Sharpe ratio, the industry’s default measure of risk-adjusted return, divides excess return by the standard deviation of returns. It treats upside volatility and downside volatility as equally undesirable. A strategy that earns 20% in one month and 0% in the next carries the same Sharpe penalty as one that earns 10% and then loses 10%. The first is a desirable profile, the second a dangerous one, and the Sharpe ratio cannot tell them apart.
For trend following this is particularly damaging. The positive skew documented throughout this series means the TF Index’s volatility sits disproportionately on the upside. The large gains that drive terminal wealth, the right-tail events of Episode 7, are penalised by the Sharpe ratio as if they were risk. The metric tells the investor to prefer smooth, moderate returns (even when the smoothness conceals left-tail risk) over choppy, positively-skewed returns (even when the choppiness is driven by occasional large gains). It penalises the very property that makes trend following geometrically valuable.
The Sharpe ratio also fails as a monitoring tool. When a trend follower captures a large crisis trend, the resulting big positive month raises the volatility of the return stream, so the Sharpe ratio declines even though the manager delivered exactly the crisis alpha that justifies the allocation. The allocator monitoring on Sharpe sees deterioration; the allocator monitoring on MAR, skew, and crisis performance sees confirmation. The metrics tell opposite stories about the same event. There is a deeper issue underneath. The Sharpe ratio treats all uncertainty as undesirable, but trend following’s uncertainty is asymmetric: the right tail is longer than the left. Uncertainty in the direction of large gains is not risk in the geometric sense. It is opportunity. The geometric framework distinguishes uncertainty that threatens the compounding base (drawdowns, left-tail events) from uncertainty that feeds it (right-tail events, crisis alpha). The Sharpe ratio collapses that distinction.
The Window Dressing Trap
The Sharpe critique has a sharper edge still. If smoothness were the signature of quality, the funds with the highest Sharpe ratios in history would be the survivors. They are frequently the casualties. Long-Term Capital Management carried a Sharpe that would have topped almost any screen right up to the quarter it failed. The systematic volatility sellers of February 2018 reported years of enviable risk-adjusted returns before a single session erased them. The pattern is not incidental. A high Sharpe ratio can be earned two ways: by producing genuinely efficient geometric returns, or by manufacturing the appearance of them. The second route, selling tail risk, levering a smooth carry, marking illiquid positions kindly, produces a beautiful Sharpe ratio precisely until it produces a catastrophe. The metric that is supposed to flag danger instead rewards its concealment.
This is a direct consequence of the market structure established in the Fractals of Finance series. Real returns are not continuous. They are fat-tailed, they cluster in volatility, and they are regime dependent. The deeper error is to treat the market as a cash-flow business at all. It is not a game that pays a steady wage for showing up. It is a game that rewards patience and the occasional exploitation of a material market movement, and those movements are neither constant nor continual. Even mean reversion, which is itself regime dependent, offers no exemption: the steady progressive ascent of the equity curve is periodically interrupted by damaging negative-skew events. That is not a flaw in the strategy. It is the nature of trading in real markets. The market does not pay in a steady drip; it pays in bursts separated by long stretches of nothing, and the bursts are where trend following earns its keep. To assume a model will generate return constantly, in every month and every regime, is not optimism. It is a claim that contradicts the structure of the markets themselves.
There is a name for the demand that performance look smoother than the underlying reality, and it comes from my old trade. In accounting we call it window dressing: the practiced art of arranging the reporting period so the figures present a calmer, more continuous picture than the business actually lived. The instinct travels straight into markets. The Window Dressing Trap is the belief that a good strategy should produce a smooth, steadily rising line, and the corollary error of preferring the strategy that merely appears to. A smooth line drawn over a wiggly market is almost always a line that has been dressed. As Alan Watts might have put it, had he ever run a book, you cannot draw a straight line through a wiggly world and call the line the truth. The wiggle is the truth. The line is the dressing.
The geometric lens refuses the dressing. It does not ask whether the path looked smooth. It asks whether the distribution compounds: positive skew, contained drawdown, crisis alpha. A trend follower’s record is honest about its wiggle, which is the very reason the Sharpe ratio undersells it and the geometric toolkit does not.
The Geometric Toolkit
The geometric framework replaces the Sharpe ratio with a set of metrics designed for the multiplicative world. Each measures a different dimension of geometric efficiency, and no single one is sufficient. The framework evaluates a manager across all seven simultaneously.
Table 1: The seven-lens geometric evaluation framework. A high MAR with negative skew may signal a process vulnerable to sudden large losses; positive skew with poor crisis performance may not deliver the portfolio diversification that justifies the patience cost. The evaluation requires all seven lenses together.
The Geometric Selection Space
The most revealing two-dimensional view of the analyzed programs plots MAR ratio against skewness, the two properties that most directly determine portfolio-level value. Figure 1 maps that selection space across the 38 long-term programs with 20-year-plus records.
Figure 1: The geometric selection space. Each program is plotted by skewness (horizontal) and MAR ratio (vertical); bubble size is proportional to terminal value. Green marks programs with positive skew and a MAR above Berkshire’s 0.240; red marks the negative-skew programs. The upper-right region holds the geometric elite: high efficiency and a favourable return shape together.
The chart reveals the structure of the population. The majority of programs cluster in the positive-skew, moderate-MAR region: the typical trend followers, geometrically sound and portfolio-enhancing, with the structural properties documented throughout the series. The upper-right region holds the geometric elite, programs that combine high efficiency with the positive skew that compounds well: Salus Alpha (MAR 0.630, skew +0.63), AHL Alpha (0.462, +0.23), AHL Evolution (0.431, skew near zero), and Graham Proprietary Matrix (0.417, +0.17).
Four programs sit in the negative-skew region: Campbell, Fort Global Contrarian, Graham K4D-10V, and Systematica BlueTrend, with skewness from roughly zero to -0.4. Negative skew in a trend-following context is a signal that warrants investigation. It may indicate that the implementation introduces left-tail risk the standard process would normally avoid, perhaps through concentrated positions, mean-reversion overlays, or insufficient diversification across timeframes. These are not necessarily bad investments, but their return shape is geometrically less favourable than the positively-skewed majority, and the evaluator should understand why before allocating.
The chart also makes a practical point. The TF Index is not directly investable. An investor accessing trend following must select individual programs or a fund-of-funds, and the selection space is the map for that decision. The investor seeking maximum geometric efficiency selects from the upper-right; the investor seeking maximum crisis alpha weights toward the strongest negative equity correlation; the investor seeking maximum standalone return accepts higher volatility and selects from the right side regardless of MAR. Each is a different answer to the same question: what geometric properties does my portfolio need?
What the Rankings Reveal
Figure 2 plots each program’s Sharpe rank against its MAR rank. For the top managers the two metrics largely agree: Salus Alpha ranks first on both, and 8 of the top 10 overlap. For practical purposes this means an allocator using Sharpe to select among the best trend followers would not make catastrophic errors. That agreement is the counterpart to the Window Dressing Trap, not a contradiction of it: a real trend follower earns its Sharpe honestly, through genuine positive skew rather than concealed tail risk, so here the two metrics point the same way. It is in the manufactured-Sharpe casualties that they diverge, and there only the geometric lens sees it coming.
Figure 2: Sharpe rank versus MAR rank across the analyzed programs. Points near the diagonal indicate agreement; bubble size is proportional to terminal wealth. The two rankings largely agree (8 of the top 10 overlap), but the divergences at the margins can be significant for allocation decisions.
The agreement, however, conceals a deeper problem. The Sharpe ratio tells the allocator nothing about why these managers rank highly. It does not reveal that their positive skew is driving terminal wealth, that their crisis performance provides portfolio diversification, or that their low volatility drag converts a high proportion of arithmetic return to geometric return. The Sharpe ratio produces the right answer for the wrong reasons, and when the reasons are wrong, the answer is fragile. An allocator who selects on Sharpe and then sees Sharpe decline may abandon a manager without understanding that the properties that actually matter, skew, crisis alpha, and drawdown containment, may be unchanged.
Consider Mulvaney. Its Sharpe ratio ranks 10th in the analyzed set, respectable but not elite, and an allocator screening on Sharpe would place it in the second tier, below programs like Winton and Quantica. Yet Mulvaney produced $75.50 per dollar, more than any other program by a factor of five. Its Sharpe is moderate because its volatility is high, at 37.4%, and the Sharpe ratio penalises all volatility equally. The geometric lens reveals a different picture: Mulvaney’s skew of +0.60 is among the highest in the database, its crisis performance is exceptional, and its arithmetic return of 23.47% is high enough to absorb a 5.44-point volatility drag. The geometric answer is that Mulvaney is a high-volatility, high-return machine that converts upside volatility into terminal wealth. The Sharpe answer is that Mulvaney is a moderately risk-adjusted strategy. One answer explains $75.50. The other cannot.
The Top Quartile
Scoring the analyzed programs across the five core geometric dimensions (MAR, skewness, Sortino, crisis performance, and equity correlation) and averaging the normalised scores produces a composite geometric score. The top quartile, the highest-scoring roughly one in four, has the following median profile.
Table 2: The top geometric quartile versus the full analyzed set and the S&P 500, Jan 2000 to Jan 2026, net of fees. The quartile’s median CAGR edges the S&P, its median MAR is roughly double the S&P’s and above Berkshire’s 0.240, and its median skew is strongly positive against the S&P’s negative skew.
These are the programs the geometric lens identifies as most valuable for portfolio construction. They do not merely compound wealth; they compound it with the distributional properties that produce the synergy documented in Episode 9: positive skew that compounds efficiently, crisis alpha that protects the compounding base, and drawdown containment that minimises the geometric drag. An allocator who selects from this quartile and combines the allocation with equities accesses the full geometric architecture.
Red Flags in Geometric Space
The toolkit does not only identify the best managers. It also flags patterns that should trigger further investigation.
Negative skew. Four of the analyzed programs exhibit negative skew. In a process designed to produce positive skew through the cut, negative skew suggests something in the implementation is introducing left-tail risk, whether intentional (a mean-reversion overlay, a concentrated position) or unintentional (insufficient diversification, correlated timeframes). It does not disqualify a manager, but it is the single most important red flag in geometric space, because it indicates a return shape that compounds less efficiently and may conceal tail risk.
High Sharpe with low MAR. A manager with a high Sharpe but a low MAR is achieving smooth returns that conceal deep drawdowns. The smoothness may come from mark-to-market practices, illiquid positions, or a process that generates steady income until it suddenly does not. Among the long-term trend followers this pattern does not appear sharply, which is itself evidence that the process does not typically produce the pathology. But in the broader universe of alternative strategies, high Sharpe with low MAR is the signature of a strategy harvesting the volatility risk premium: profitable most of the time, catastrophic in the tail.
Extreme volatility drag. Episode 10 documented Superfund Green SPC 2x, with a 7.03-point drag, as the most dramatic case of arithmetic return eroded by the volatility tax. High drag is not inherently bad if the arithmetic return is high enough to absorb it (Mulvaney’s 5.44-point drag still leaves an 18.03% CAGR). But for a manager with a moderate arithmetic return, high drag warns that the volatility target may be too aggressive for the strategy’s capacity to generate return.
The Picks Illusion
The investment industry is built on a founding narrative: wealth comes from picking the right assets. The best stock picker wins, the best sector allocator wins, the best macro trader wins, and evaluation frameworks are designed to measure picking ability through alpha, information ratio, hit rate, and selectivity. The geometric lens inverts this narrative. The data in this series shows that systematic trend-following programs, which make no picks at all, beat the greatest stock picker in history on geometric efficiency, with seventeen of the analyzed programs exceeding Berkshire on MAR. They do so not by selecting better assets but by producing a return stream with better geometric properties: positive skew, drawdown containment, and crisis alpha. The picks are inputs. The geometry is the output. The output is what compounds.
The standard question is “What did they pick?” The geometric question is “What are the properties of the return stream?” The picks are inputs. The geometry is the output. Evaluate the output.
This has direct implications for evaluation. A manager with brilliant picks and negative skew will compound less wealth than a manager with unremarkable picks and positive skew. A manager with high alpha and a 50.9% maximum drawdown will spend years recovering instead of compounding. The framework should weight geometric properties above selection metrics, because the geometric properties are what determine terminal wealth. This does not mean skill does not exist or does not matter. It means skill is best measured by the geometric properties of the output rather than the apparent quality of the inputs. A manager who consistently produces positive skew, contained drawdowns, and crisis alpha is demonstrating skill, even if no individual position is inspired. The skill lies in the construction of the process, not the brilliance of the trades.
The Practical Framework
The following sequence applies the geometric toolkit in practice. It is sequential, not formulaic: the investor applies each lens in order, narrowing the pool and building an understanding of each remaining manager’s geometric architecture.
Step 1: Screen on skewness. Eliminate or flag any manager with persistent negative skew. The remaining pool contains the managers whose process produces the return shape that compounds efficiently.
Step 2: Rank on MAR. Among the positively-skewed managers, rank by MAR. This identifies those extracting the most compounding from the least geometric risk.
Step 3: Verify crisis performance. Check returns during the GFC, 2022, and other equity-stress periods. A manager with a high MAR but poor crisis performance may be achieving efficiency through low volatility rather than the crisis alpha that provides portfolio value.
Step 4: Check volatility drag. Compare arithmetic return to CAGR. A drag above three percentage points means a substantial tax on volatility; assess whether the arithmetic return justifies it.
Step 5: Assess equity correlation. Confirm negative or near-zero correlation to the S&P 500. This is the property that delivers the portfolio synergy of Episode 9. A trend follower with positive equity correlation provides returns but not diversification.
The goal is not a single score. It is to understand why a manager compounds at the rate it does, on the path it takes, with the distributional properties it exhibits, because that understanding is what allows the investor to hold through the patience cost of Episode 12. Two further points apply. For managers with shorter records, skewness, MAR, and volatility drag are measurable over any period, but crisis performance requires a crisis to manifest; a five-year record entirely within a bull market has not yet been tested under stress, and the evaluator should weight the other lenses more heavily while recognising the most important test has not yet occurred. And all evaluation should be conducted on net-of-fee returns, because a manager whose geometric efficiency is high gross but mediocre net is consuming the edge in fees, and the investor receives only the geometry that remains after fees.
The Geometric Lens on Other Strategies
The toolkit was developed here to evaluate trend following, but MAR, skewness, volatility drag, and crisis performance are properties of any return stream. The strongest test of a framework’s validity is whether it produces coherent results outside the domain it was built for. Applying the same seven lenses to the three most prominent alternatives, value investing, volatility selling, and carry, does exactly that.
Value investing, long-only equity selection on fundamental cheapness, is the canonical alternative to systematic process, and its geometric profile is well documented. The return distribution is negatively skewed: sharp, sudden drawdowns during panics (when cheap stocks get cheaper) and slow recovery through steady accumulation. The Fama-French HML value factor has historically shown skewness around -0.3 to -0.5, and the factor’s drawdown during the GFC exceeded 50%. Most importantly, value’s crisis correlation is typically positive: it loses money when equities do, often more severely, because the mechanism that produces cheap stocks (fundamental deterioration) accelerates in recessions. The geometric lens identifies value as a negatively-skewed, positively-correlated strategy with a low MAR and high drag. It earns a real premium, but the return stream is geometrically hostile: it concentrates variance on the left and performs worst when the investor can least afford it.
Volatility selling, the systematic collection of option premium, is the strategy that superficially resembles the geometric ideal. Its win rate is high, since implied volatility usually exceeds realised, and its Sharpe in calm regimes can look impressive. But the geometric lens sees the structural deception immediately. Volatility selling has extreme negative skew: the premium-collection months are small, and the loss months, when volatility spikes, are catastrophic. Benchmarks for systematic put-selling recorded a drawdown around 32% during the GFC and a near-instantaneous 25% loss during the February 2018 volatility spike. The MAR collapses, and the Sharpe ratio cannot detect the pathology because it treats the smooth gains and the catastrophic tail as symmetric. High Sharpe, low MAR, negative skew: precisely the red-flag pattern. Volatility selling is the strategy the Sharpe ratio over-rewards and the geometric toolkit correctly flags.
Carry, being long high-yielding assets and short low-yielding ones, occupies a middle ground. Its long-run premium is well established, with diversified carry historically showing a Sharpe around 0.4 to 0.6. But the skewness is negative: carry suffers sudden, severe reversals during risk-off events, when funding currencies appreciate sharply and force rapid unwinds across a correlated book, as in 2008 and the yen-carry reversal of mid-2024. Its crisis correlation with equities is positive and often strongly so, removing the diversification that would justify holding it alongside equities. The geometric lens identifies carry as a reasonable-Sharpe, reasonable-CAGR strategy with negative skew, positive equity-crisis correlation, and periodic deep drawdowns: it diversifies the calm but not the storm.
Value concentrates variance on the left. Volatility selling creates a smooth path to a catastrophic tail. Carry diversifies the calm but not the storm. The geometric lens reaches the same conclusion in each case: a high Sharpe ratio is not a reliable guide to geometric wealth. The shape of the distribution is what compounds.
This is not an argument that value, volatility selling, or carry are bad strategies; each earns a documented premium. The argument is that the geometric lens is the correct lens for all of them, and that applied consistently it reveals what the Sharpe ratio obscures: value’s negative skew, volatility selling’s catastrophic left tail, and carry’s positive crisis correlation are all invisible in a Sharpe ratio and immediately visible in skewness, crisis performance, and MAR. The framework is not purpose-built to favour trend following. It is purpose-built to reveal the geometry of whatever return stream it examines. In this case, that geometry happens to distinguish trend following from its major alternatives on the dimensions that determine terminal wealth.
The Toolkit in Action
Theory is necessary but insufficient. A toolkit that ranks managers correctly on historical data but fails to select good future compounders is an academic exercise. The strongest test of an evaluation framework is whether it produces superior outcomes applied strictly out of sample, using only information available at the time of each decision, with no optimisation, no parameter fitting, and no retrospective adjustment. The following walk-forward test applies exactly that standard.
The discipline is absolute. The investable universe in any given year is the full set of long-running trend followers that have accumulated at least a 15-year audited track record as of that date, a pool of more than fifty programs drawn from the complete database, including programs that later closed. At the end of each year the allocator looks back over a rolling 15-year window, evaluates candidate ten-manager ensembles by MAR, selects the highest-MAR ten-manager composite, equal-weights it, and holds for the following calendar year. At each year-end the window rolls forward one year, the oldest year drops away, and the selection repeats. No parameters are optimised. No future data contaminates any selection. Every decision uses only information that was available on the day it was made.
The scale of the selection problem is worth stating. Choosing the best ten-manager ensemble from a universe of fifty spans more than ten billion possible combinations. An exhaustive optimisation using the entire sample at once would find the single best combination in hindsight, but that information is not available at the moment decisions must be made: the past is complete, the future is not. The walk-forward process cannot see the future, so instead of finding the hindsight-optimal ensemble it approximates it, year by year, using only what has already happened. The question is how close a disciplined, foresight-free process can come to the unreachable hindsight ceiling.
It comes remarkably close. Traded forward through time, the walk-forward ensemble compounds at roughly 7.1% annually with a maximum drawdown near 11.5% and positive skew, for a MAR in the region of 0.6. Table 3 sets that against the benchmarks over the same window.
Table 3: The walk-forward ten-manager ensemble against the benchmarks, net of fees. The ensemble trails the S&P modestly on raw CAGR but more than doubles the TF Index’s MAR and nearly quadruples the S&P’s, with a maximum drawdown well under a quarter of the S&P’s. The result is achieved walk-forward, with no look-ahead at any selection.
Read the table through the geometric lens rather than the CAGR column alone. The ensemble’s maximum drawdown of about 11.5% is roughly a fifth of the S&P’s 50.9% and around half the TF Index’s, and its MAR sits near 0.62 against the S&P’s 0.157. That efficiency, not the headline return, is the point: a disciplined, foresight-free selection process built only on geometric properties produces a return stream that compounds smoothly and contains its drawdowns far better than either benchmark.
Two features of the process deserve emphasis. First, the selections rotate naturally among long-term survivors. No single manager is permanent; the ensemble adapts each year as the rolling window evolves, and the rotation is a map of diversification through time rather than a prediction. Second, the walk-forward result converges close to the hindsight-optimal benchmark precisely because the geometric properties are real, persistent, and measurable. If they were artefacts of a particular window or regime, the rankings would churn endlessly and the forward results would disappoint. They do not. The process reaches nearly the same destination as hindsight, but it does so by walking forward through time, never looking back.
This is an institutional allocator’s method. The selected programs typically require minimum investments of one million dollars or more, so a retail investor cannot replicate it directly. But the principle is accessible through a diversified composite or fund-of-funds: the allocator receives the geometric fingerprint, positive skew, crisis alpha, and negative equity correlation, while diversifying away a substantial part of individual-manager risk. This is the ensemble benefit of Episode 11, applied as a practical selection tool, and it is the strongest possible answer to the hindsight-bias objection. The objection says: you are showing us the winners and then explaining why they won. The walk-forward test inverts it, showing that a mechanical process applied with strict temporal discipline identifies durable compounders before the fact, with no human judgement, no narrative, and no market view.
The Bridge
The toolkit is now complete, and it has been tested. The investor knows the mathematics (Episode 2), the mechanism (Episodes 5 through 8), the portfolio architecture (Episode 9), the evidence (Episodes 10 and 11), the cost (Episode 12), and the evaluation framework (this episode). The walk-forward test demonstrates that the toolkit does not merely describe the past; it selects managers that compound wealth going forward. Episode 14 will assemble these pieces into a complete portfolio-construction guide for the multiplicative world: how much to allocate, how to rebalance, how to implement, and how to monitor a geometric portfolio over time. The series moves from evaluation to construction.
Data and Sources
All performance data is drawn from the NilssonHedge Trend Following performance file, net of management and performance fees. The cross-sectional exhibits (Figures 1 and 2 and Table 2) use the 38 long-term programs with 20-year-plus records over the window January 2000 to January 2026, the same horizon-consistent subset used in Episodes 10 and 11; the wider analysis universe is the 47-program TTU TF Index defined by a minimum 15-year track record. Sharpe ratio is computed as (CAGR minus 2%) divided by annualised volatility. Sortino ratio is (CAGR minus 2%) divided by annualised downside deviation, using only negative monthly returns. MAR is CAGR divided by the absolute maximum drawdown. The composite geometric score normalises MAR, skewness, Sortino, GFC performance, and negative S&P correlation to a zero-to-one scale and averages them; the top quartile is the highest-scoring roughly one in four.
The walk-forward allocation test uses a rolling 15-year window and a point-in-time eligible universe of more than fifty long-running programs, each with at least a 15-year audited track record as of the selection date, drawn from the complete database including programs that subsequently closed. Each year the highest-MAR ten-manager ensemble is selected, equal-weighted, and held for the following calendar year, after which the window rolls forward and the selection repeats. No parameters are optimised and no hindsight data is used; all reported returns are live, realised, and net of fees. The walk-forward ensemble compounds at approximately 7.1% annually with a maximum drawdown near 11.5%.
The Sharpe-ratio critique draws on Taleb, “Statistical Consequences of Fat Tails” (2020), and Lo, “The Statistics of Sharpe Ratios” (Financial Analysts Journal, 2002). The comparative figures for value investing, volatility selling, and carry derive from external academic and index literature and are not reproducible from the performance database; they should be cited to their original sources when this episode is published.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets
The book explores the full architecture of feedback, fat tails, and fractal structure in financial markets, and what it means for how we trade, invest, and understand risk.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.