A dollar invested in Mulvaney’s Global Diversified Program in January 2000 is worth $75.50 today. A dollar in the S&P 500 is worth $7.48. A dollar in Berkshire Hathaway is worth $14.11. The geometry of wealth is the difference between $75.50 and $7.48.
Episode 9 demonstrated that blending equities with trend following produces geometric synergy: more terminal wealth and less drawdown than either component alone. That analysis used the TTU TF Index, a composite of many programs. This episode opens the composite and examines the individual machines inside it.
This is the beginning of Act III. The previous nine episodes built the theoretical case: the mathematics of compounding, the mechanism of trend following, and the portfolio architecture that turns these properties into wealth. Act III deploys the data. It asks whether the theoretical case holds against decades of independent track records, it confronts the objections, and it begins with the most basic question of all: what did the compounding machines actually produce?
The TTU TF Index comprises 47 systematic, globally diversified trend-following programs, each meeting the Index’s inclusion standard of a minimum 15-year unbroken track record to the current reporting month. To compare programs on a like-for-like basis over the full evidence window, the per-program exhibits in this episode draw on the 38 of those programs whose records span at least 20 years. That is a record long enough to have been tested through the global financial crisis, and for the long-tenured majority, the dot-com bust as well. Each program is measured from January 2000, or its inception within the window, to its latest available month.
The answer ranges from extraordinary to modest, and both ends carry information. The extraordinary outcomes show what the process can produce when calibration, markets, and endurance align. The modest outcomes show how sensitive terminal wealth is to the volatility target and to the geometric drag that Episode 2 warned about. Together they tell the full story of what a quarter-century of trend-following compounding actually looks like.
The Equity Curve Gallery
Figure 1 shows the cumulative growth of $1 invested in the five highest-performing trend followers in the database, alongside the three benchmarks that have anchored every episode: the S&P 500, Berkshire Hathaway, and the TTU TF Index.
Figure 1: Cumulative equity curves (log scale) for the top five trend followers by terminal value, alongside the S&P 500, Berkshire Hathaway, and the TTU TF Index (Jan 2000 to Jan 2026). Shaded regions mark equity crises. Mulvaney’s curve dominates, but its path includes a 60.9% drawdown. Each curve begins at $1 at its inception within the window.
Mulvaney’s Global Diversified Program turned $1 into $75.50 over 313 months, a CAGR of 18.03%. This is the highest terminal value in the database by a wide margin and exceeds Berkshire Hathaway’s $14.11 by a factor of five. The comparison is instructive. Berkshire is widely regarded as the most successful investment vehicle in modern history. Yet a systematic trend-following program, operating without fundamental analysis, without any view on the economy, and without human judgement about which positions to hold, produced five times the terminal wealth over the same period. The geometry did the work. Mulvaney’s outcome is driven by aggressive pyramiding into winning trends at high leverage across a concentrated set of roughly 45 markets, not by the broad diversification that the Wide Net section identifies as the structural wealth driver. The consequences of that approach, including a 60.9% maximum drawdown, are examined in the Leverage Fallacy section.
Salus Alpha’s Directional Markets turned $1 into $16.38, a CAGR of 13.03% with a maximum drawdown of just 20.7%. Its MAR of 0.630 is the highest in the database: it extracts more compounding per unit of drawdown than any other machine. Lynx, Winton Diversified Macro, and Winton Multi-Strategy each turned $1 into roughly $8.50, broadly comparable to the S&P 500 but with markedly different paths: shallower drawdowns, positive skew, and the crisis alpha documented in earlier episodes.
During every equity crisis, when the S&P 500 and Berkshire fall sharply, most of the trend-following curves rise or hold flat. Mulvaney surges through the dot-com bust and the GFC. Salus Alpha compounds steadily through every crisis without the violent drawdowns visible in the equity benchmarks. The crisis alpha documented in Episode 8 is visible in every individual equity curve, not as a statistical property of the composite but as a lived characteristic of each independent machine.
The range is wide. Not every program produces extraordinary terminal wealth; the bottom of the database includes a program that turned $1 into just $1.32, barely outpacing cash. The full range reflects the diversity of implementations: different volatility targets, market universes, lookback periods, and position-sizing rules. The process is the same. The calibration is not. And in a multiplicative world, small differences in calibration compound into enormous differences in outcome.
The Full Record
Table 1 presents the top ten programs by terminal value alongside the three benchmarks. It reveals the geometric architecture of each compounding machine.
Table 1: Top ten programs by terminal value (growth of $1, Jan 2000 to latest available month) with three benchmarks. Salus Alpha is measured over 274 months (from Mar 2003) and AHL Evolution over 244 months (from Oct 2005); CAGR and MAR are annualised and comparable, but terminal wealth reflects each program’s own horizon.
Eight of the analyzed programs produced a CAGR exceeding the S&P 500’s 8.02%. Seventeen produced a MAR exceeding Berkshire’s 0.240. That is the more significant finding: roughly two in five of the database delivered superior geometric efficiency to arguably the greatest equity investor in history. They did so not by earning higher raw returns but by managing the path more efficiently.
Table 1 also reveals two archetypes. The first is the concentrated right-tail exploiter, exemplified by Mulvaney: an arithmetic return of 23.47% generated by pyramiding aggressively into winning trends at high leverage, large enough to overwhelm the volatility drag even at 37.4% annualised volatility, producing 18.03% CAGR at the cost of a 60.9% drawdown. The second is the geometric optimizer, exemplified by Salus Alpha and Man AHL Alpha: lower arithmetic returns across broader universes, but converted almost entirely to geometric return because the paths are smoother. Salus Alpha’s drag is just 0.14 percentage points; AHL Alpha’s is 0.25. They do not sacrifice return for safety; they are structured so geometric efficiency and return reinforce each other.
The Geometric Decomposition
The final two columns of Table 1 reveal the mechanism. The arithmetic return is what the strategy earns on average each year, treating each year as independent. The volatility drag is the gap between arithmetic and geometric return, the tax that compounding imposes on volatile return streams. Every strategy pays this tax. The question is how much.
Figure 2: Geometric decomposition for the 38 trend programs with a 20-year-plus record and the three benchmarks (Jan 2000 to Jan 2026). Green bars show CAGR (geometric return kept); coral bars show the volatility drag subtracted from arithmetic return. Programs with small coral bars convert more of their arithmetic return into geometric wealth. Benchmarks shown in bold.
Salus Alpha illustrates the efficient case: an arithmetic return of 13.17% with a drag of just 0.14 percentage points. Nearly all of it converts to geometric return, which is why its MAR of 0.630 is the highest in the database. DUNN WMA illustrates the opposite: a high arithmetic return of 11.87%, but a 29.3% volatility imposing a 3.95-point drag, leaving a 7.92% CAGR. DUNN earns substantially more per year on average than the S&P 500 yet compounds to less, because the path is rougher.
The TTU TF Index itself demonstrates the efficiency. Its arithmetic return of 8.16% is lower than the S&P’s 8.89%, but its drag of 0.61% is smaller than the S&P’s 0.88%, which is why it achieves a MAR of 0.359 versus the S&P’s 0.157. Geometric efficiency is not about earning more. It is about keeping more of what you earn. In Figure 2 the green bars show what each strategy keeps; the coral bars show what it loses. The smallest coral bars belong to the geometric optimisers. This is the quantitative expression of Episode 2’s core argument: two strategies with identical arithmetic returns produce different terminal wealth if their paths differ, and the smoother path wins.
The Leverage Fallacy
A sharp reader will have noticed something. Salus Alpha’s MAR of 0.630 implies that if leveraged 2x it would achieve roughly 26% CAGR at a 41% drawdown, demolishing both Mulvaney and Berkshire. If MAR measures geometric efficiency, and if leverage scales return and drawdown proportionally, then the optimal strategy is to find the highest-MAR program and lever it up.
This reasoning is arithmetically correct and geometrically catastrophic. It is the leverage fallacy, and it rests on an assumption that sounds reasonable but is false: that CAGR and maximum drawdown scale the same way with leverage. They do not. CAGR is path-dependent; maximum drawdown is a single point observation that scales roughly linearly with leverage. But the volatility tax from Episode 1 applies with devastating force: the tax is approximately one half of the variance, and variance scales with the square of leverage. Double the leverage, and the tax at least quadruples.
The data proves it. Table 2 applies increasing leverage to the TTU TF Index over the full record. Every row uses the same underlying return stream; the only variable is the leverage multiple.
Table 2: TTU TF Index at increasing leverage (Jan 2000 to Jan 2026). CAGR scales sub-linearly; drawdown scales roughly linearly; the volatility tax scales quadratically; MAR falls. The “$1 Becomes” column is computed directly from each row’s CAGR over 313 months.
Read Table 2 carefully. At 2x leverage the volatility tax rises from 0.61% to 2.62%, a factor of 4.3. The drawdown rises from 21.0% to 38.4%, a factor of 1.83. But the CAGR rises from 7.55% to only 13.69%, a factor of 1.81. Leverage delivered 1.81 times the CAGR for 1.83 times the drawdown and 4.3 times the geometric drag. At 3x the tax consumes 6.29 percentage points per year and the drawdown breaches 50%. The MAR, which the leverage argument assumes is constant, falls from 0.359 to 0.345. Leverage degrades geometric efficiency. It does not preserve it.
The Mulvaney case is starker, and Figure 3 makes it visible. Mulvaney already runs at 37.4% volatility. At 2x leverage the maximum drawdown collapses past 90%, the volatility tax explodes beyond 26 percentage points per year, and the CAGR rises only marginally before the tax overwhelms it. The 1.5x terminal value ($170) actually exceeds the 2x terminal value ($125). No investor sustains a 90%-plus drawdown without abandoning the process. The 18.03% CAGR Mulvaney delivered through a 60.9% drawdown is already at the limit of human endurance; doubling the drawdown does not double the CAGR, it barely moves it, and it eliminates the investor.
Figure 3: The leverage fallacy in two panels. Panel A: the TTU TF Index at 1x, 1.5x, 2x, 3x. The volatility tax at least quadruples at 2x while CAGR scales sub-linearly. Panel B: Mulvaney at 0.5x, 1x, 1.5x, 2x. At 2x the drawdown reaches 93% for only 2.4 points of additional CAGR, and the 2x curve finishes below the 1.5x curve. CAGR is path-dependent; drawdown is not. Leverage degrades the geometry.
This is not a failure of leverage in practice but in pure mathematics. The half-variance relationship is exact in the continuously-compounded (log-return) limit and a close approximation otherwise; for fat-tailed return streams the realised drag tends to run slightly above the pure quadratic, so “at least quadruples” is the honest statement. Even in a frictionless world with no margin calls and no liquidity costs, the volatility tax ensures that doubling leverage cannot double CAGR. Every practical friction (margin requirements that rise in crises, slippage that scales non-linearly, the psychological impossibility of a 90% drawdown) acts in the same direction, widening the gap between levered arithmetic expectation and levered geometric reality.
The leverage fallacy reveals a deeper truth. CAGR is the geometric growth rate of a multiplicative process: a property of the entire path, encoding the sequence of returns, the depth of drawdowns, and the cumulative effect of every month of volatility drag. Maximum drawdown is a single point observation. You can scale a point observation and predict the result; you cannot scale a path-dependent quantity and predict the result, because the path itself changes under leverage, and the volatility tax ensures it changes for the worse. Any investor who believes they can take a high-MAR strategy, apply leverage, and manufacture a high-CAGR strategy is committing the same error as the investor who believes the arithmetic average represents their wealth.
The Wide Net
The leverage fallacy is one way investors try to manufacture geometric wealth. The narrow portfolio is another, and equally destructive. Financial markets are complex adaptive systems that produce fractal geometry: extreme events, the outliers, are not statistical accidents but structural inevitabilities, appearing at every scale. The question is never whether an outlier will occur, but which market, which month, and in which direction.
This reframes diversification. The standard argument is that it reduces risk by smoothing variance. In a system where outliers are structurally omnipresent but individually unpredictable, diversification does something more powerful: it removes the need to be lucky. At any moment, most positions in a trend-following portfolio are noise, small bets in markets oscillating sideways. A few, perhaps five to ten percent, are catching a structural trend: a supply shock, a policy divergence, a repricing. Those are the trunks. They are the outliers that drive terminal wealth.
An investor trading twenty markets has twenty independent draws from the fat-tailed distribution; one trading two hundred has two hundred. The probability that at least one market produces a trunk-scale outlier in a given period rises with the number of draws. This is an order-statistics result, but it must be stated correctly: for the fat-tailed (power-law) distributions that describe real markets, the expected maximum of N independent draws grows as a power of N, approximately N raised to one over the tail index, with that index near three for financial returns. That growth is sub-linear in N, yet it is dramatically faster than the painfully slow growth of the Gaussian maximum, which rises only with the square root of the logarithm of N. Each additional market therefore enlarges the single biggest outlier the portfolio can catch far more than Gaussian intuition suggests, not because individual markets become more volatile, but because a wider net intersects a system where extreme events are always occurring somewhere.
Each captured outlier does not merely add to the portfolio; it multiplies the compounding base. When a trend-following portfolio captures a twenty-percent gain from a cocoa supercycle or a bond repricing, that gain permanently elevates the capital from which all future compounding proceeds. Miss it because the market was not in your universe, and you miss not only the twenty percent but every dollar it would have compounded into over the remaining life of the portfolio. In a multiplicative system, the cost of a missed outlier is the entire compounding chain that would have followed.
This produces a distinctive equity curve: long flat stretches punctuated by sharp upward steps when a position catches a structural trend. Each step is an irreversible ratchet in the compounding base. The stepped curve out-compounds the smooth curve of equal arithmetic average because each step is a multiplication, not an addition. Cocoa in 2024, crude oil in 2020, natural gas in 2022, the Japanese yen in 2024: each produced trunk-scale moves that were individually unpredictable but collectively inevitable. Programs holding those markets captured the steps; programs with narrower sets waited for outliers that never arrived in their universe. A narrow portfolio catches twigs. A broad portfolio catches trunks.
The Range and What It Means
The spread between the best and worst programs is enormous: Mulvaney’s $75.50 against a floor barely above cash, a difference of more than fiftyfold. Part of this is the leverage fallacy in reverse: programs running at excessive volatility pay a quadratic drag that consumes their arithmetic return. The Superfund Green SPC line illustrates the cost of volatility, though a note of caution applies to reading its share classes as a clean leverage ladder: the 1x, 1.5x and 2x classes in the database cover different start dates and different numbers of months, so they are not a controlled like-for-like leverage experiment and should not be read as one. Table 2 (a single index, a single window, scaled mathematically) is the controlled demonstration.
The programs at the bottom of the database are not failed implementations of the process. They chose different volatility targets, market universes, or position-sizing rules. The process worked in all cases, producing the characteristic positive skew and crisis alpha documented in Episodes 7 and 8. Terminal wealth varies enormously because small differences in geometric drag and universe breadth compound into vast differences over decades.
A sceptic will raise survivorship bias, and it deserves an honest answer. The inclusion rule is a minimum 15-year unbroken track record, so every program analyzed here has endured at least one full cycle of trend-following feast and famine, including the prolonged 2015 to 2020 drought and the 2022 crash that broke many discretionary strategies. The 38 programs in the per-program exhibits each carry a 20-year-plus record; every one of them lived through the global financial crisis, and the long-tenured core, those trading since January 2000, endured the dot-com bust as well. Survival through those regimes is itself evidence that the process does not self-destruct under stress. More important, the properties that matter for portfolio construction (positive skew, crisis alpha, and negative equity correlation) are present across the full range of performers, not just the top. Even the weakest programs by terminal value show positive skew and negative equity correlation. The geometric properties are structural. The terminal wealth is variable.
It is worth being explicit about the 20-year cut used for the per-program exhibits: it is a track-length criterion fixed in advance, not a performance screen. No program was included or excluded on the basis of its returns, only on the length of its record, so the exhibits cannot be accused of selecting for the result they display.
The Running Ledger
Our running comparison now includes the individual programs that anchor the gallery.
The Bridge
This episode presented the raw compounding record: dozens of machines, a quarter-century of data, outcomes ranging from barely-above-cash to $75.50. It identified two archetypes: the concentrated right-tail exploiter whose arithmetic firepower overwhelms the geometric drag, and the geometric optimizer whose diversification and path discipline convert a lower arithmetic return into superior risk-adjusted compounding. It confronted the leverage fallacy and showed that CAGR cannot be manufactured by scaling a MAR ratio. And it established that maximum diversification is a geometric strategy, not merely a risk-management preference, because the coupled market system generates structurally inevitable tail events continuously across the full universe of markets.
But the record raises a question the individual equity curves cannot answer. One program’s excellence might be skill. A handful might be coincidence. When roughly seven in eight of the analyzed programs produce positive skew, about three in four produce negative equity correlation, and two in five produce a MAR exceeding Berkshire, the pattern demands a different explanation: not individual brilliance, but collective convergence.
Episode 11 will examine that convergence. It will show that consistency across dozens of independent implementations is itself the strongest evidence that the geometric properties documented in this series are structural rather than personal.
Data and Sources
All performance data is drawn from the NilssonHedge CTA database as captured in the TTU Trend Following performance file for the period January 2000 to January 2026 (313 months). Individual program returns are taken from the CrossTab Dump sheet in decimal form and converted to percentages for analysis. The analysis universe is the TTU TF Index: 47 active, systematic, globally diversified trend-following programs, each meeting the Index’s inclusion standard of a minimum 15-year unbroken track record to the current reporting month. All returns are net of management and performance fees.
Every program is measured over the common window from January 2000 (or its inception within that window) to its latest available month. Programs therefore differ in the number of months they contribute: those active since January 2000 span the full 313 months, while later entrants span fewer (for example, Salus Alpha from March 2003 and AHL Evolution from October 2005). CAGR, volatility, MAR, skew, and the volatility drag are annualised or otherwise horizon-normalised and are directly comparable across programs; growth-of-$1 terminal values are not, since they reflect each program’s own number of compounding months, and are labelled accordingly. To keep the cross-program exhibits (Figures 1 and 2, Table 1) on a consistent horizon spanning the full evidence period, those exhibits use the 38 programs with a 20-year-plus record.
CAGR is computed as (terminal)^(12/months) − 1. Volatility is the annualised standard deviation of monthly returns. Maximum drawdown is the largest peak-to-trough decline in the cumulative return series. MAR = CAGR / |MaxDD|. Arithmetic return is the annualised mean of monthly returns. Volatility drag = arithmetic return − CAGR. The leverage analysis (Table 2, Figure 3) applies the leverage multiple directly to the monthly returns of the TTU TF Index and of individual programs; the resulting CAGR, volatility, and drawdown are computed from the levered monthly series. This is a mathematical demonstration of the quadratic scaling of the volatility tax and does not account for practical frictions (margin, liquidity, correlation-regime changes), which would further degrade levered outcomes.
Benchmark series (S&P 500 Total Return, Berkshire Hathaway, and the TTU TF Index) are taken from the Data sheet of the same file over the identical window. All benchmark and headline-program figures in this episode were verified directly against the source monthly returns.
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.