Why Evolution Picks Trend-Following Over Buy-the-Dip

The debate between trend-following and mean-reversion is as old as systematic trading itself. Momentum traders argue that prices exhibit persistent directional moves. Mean-reversion advocates counter that prices always return to fair value. Both camps have academic support, backtests, and decades of live results to point to. What neither side has had, until recently, is an evolutionary experiment: put both strategies in a survival-of-the-fittest arena and see which one is left standing after thousands of generations of natural selection.

Chen (2026) does exactly that. The paper constructs a simulation called MAS-Utopia, populates it with 10,000 agents running five distinct trading archetypes, feeds them five years of high-frequency market data, and lets evolution take its course. The result is unambiguous.

The setup: five archetypes, zero excuses

The simulation's design is deliberately idealized to isolate strategy fitness from structural advantages. There are no transaction costs. Every agent that goes bankrupt gets reincarnated with a fresh id="post-body" class="prose prose-lg max-w-none pb-16" style="font-family: 'Inter', sans-serif; line-height: 1.8; color: #374151;" data-astro-cid-xnoknld4>0,000 in capital (an unconditional basic income mechanism). This removes the argument that some strategies only fail because of fees or undercapitalization. Whatever survives in this environment survives because of its mathematical properties, not because of luck or structural tailwinds.

The five archetypes represent the main heuristics that retail and institutional traders actually use:

Trend-Following aligns with the prevailing market direction. It buys rising assets and sells falling ones, using macro-regime filters and confirmation signals. Low trade frequency, moderate leverage.

Mean-Reversion bets against the current, buying oversold conditions and shorting overbought ones. The classic "buy the dip" logic. It profits when prices oscillate around a mean.

Grid Trading places orders at regular intervals above and below the current price, profiting from oscillations. Similar to mean-reversion in its assumption that prices range, but more mechanical.

Sniper uses extreme leverage on high-conviction setups, betting large on single opportunities. Low frequency, high risk per trade.

HFT (High-Frequency) trades constantly, attempting to capture microstructure patterns through volume and speed.

Each agent carries a genetic vector encoding its risk parameters: leverage, position size, stop-loss width, take-profit levels, and cognitive bias factors (FOMO, panic). When an agent goes bankrupt (maximum drawdown exceeds 40%), it is replaced by a new agent that either inherits traits from successful survivors or is randomly generated. This creates genuine Darwinian selection pressure.

The outcome: trend-following takes over

After five simulated years with over 520,000 time steps, the population distribution tells a clear story:

Trend-Following: 7,659 survivors (76.6%), avg. ROI +14.71%, avg. 312 trades

Mean-Reversion: 865 survivors (8.7%), avg. ROI +0.38%, avg. 1,483 trades

Grid: 1,030 survivors (10.3%), avg. ROI -4.69%, avg. 1,254 trades

Sniper: 233 survivors (2.3%), avg. ROI -3.91%, avg. 1,665 trades

HFT: 213 survivors (2.1%), avg. ROI -0.62%, avg. 1,277 trades

Trend-following did not merely outperform. It colonized the population. Starting from an equal 20% share, it grew to over 76% of all surviving agents. Mean-reversion survived but barely broke even. Grid, Sniper, and HFT were functionally eliminated.

The most striking detail: trend-following achieved the highest average return with the lowest trade count. The agents that traded least frequently and only aligned with macro-directional moves were the ones that survived.

Why mean-reversion is structurally fragile

The mechanism is not complicated, but it is often misunderstood by practitioners who focus on win rates rather than return distributions.

Mean-reversion strategies have negatively skewed returns. They produce many small, consistent profits when markets oscillate within a range. But when a strong trend develops, the mean-reversion trader is on the wrong side. The losses during trending phases are not just large, they are catastrophic relative to the small gains accumulated during ranging periods. This is mathematically analogous to selling out-of-the-money options: consistent premium income punctuated by occasional blowups.

Trend-following has the opposite profile. Its returns are positively skewed. It endures many small losses during choppy, directionless markets (whipsaws), but captures outsized gains when genuine trends emerge. Fat-tailed market events, the kind that destroy mean-reversion accounts, are exactly what trend-following profits from.

In the simulation, the periodic occurrence of strong trending phases acted as an evolutionary filter. Each major trend wiped out a cohort of mean-reversion agents. Each wipeout shifted the population further toward trend-following. Over enough iterations, this asymmetric selection pressure produced near-total dominance.

The survival zone

The simulation also reveals which parameter combinations lead to survival. Profitable agents cluster overwhelmingly in the low-leverage (below 10x) and moderate stop-loss zone (1 to 4 ATR multiples). Agents operating at high leverage, regardless of their strategy type, exhibit almost universally negative returns.

This is a mathematical inevitability that the paper formalizes: at 100x leverage, the liquidation distance is 1% from entry. At 5x leverage, it is 20%. The higher the leverage, the tighter the margin for error, and the higher the probability that random market noise will trigger forced liquidation before a thesis plays out. The evolutionary winners were not the agents that maximized return per trade but the ones that maximized their probability of surviving long enough to participate in the next big move.

Limitations worth noting

The paper has genuine strengths in its experimental design, but a few caveats are relevant for anyone drawing practical conclusions.

First, the simulation uses cryptocurrency market data (100 assets, 5-minute bars). Crypto markets are structurally more trending and more volatile than equity or fixed-income markets. The dominance of trend-following may be amplified by this choice of data. In calmer, more mean-reverting markets (like short-term bond yields or equity pairs), the balance might shift.

Second, there is no endogenous price formation. Agent actions do not affect prices. In real markets, the success of trend-following at scale creates its own feedback loops (crowding) and eventual reversal patterns. The simulation cannot capture this.

Third, the paper is a preprint by an independent researcher. It has not yet undergone peer review. The methodology is transparent and the code is described in detail, but external validation is still pending.

Fourth, the UBI mechanism guarantees infinite reincarnation. In real markets, a blown account stays blown. The selection pressure is actually stronger in reality than in the simulation, which if anything makes the case for trend-following more conservative, not less.

What this means for systematic investing

The paper does not prove that trend-following always works or that mean-reversion never works. Markets go through extended ranging periods where trend-following bleeds and mean-reversion thrives. What the evolutionary framing shows is something more fundamental: over a sufficiently long horizon with fat-tailed return distributions, strategies with positive skew (trend-following) have a structural survival advantage over strategies with negative skew (mean-reversion).

This aligns with decades of empirical evidence. Moskowitz, Ooi, and Pedersen (2012) documented time-series momentum across 58 liquid instruments spanning equity indices, currencies, commodities, and bonds. Jegadeesh and Titman (1993) showed cross-sectional momentum in equities persisting for 3 to 12 months. The Carhart (1997) four-factor model added momentum as a systematic risk factor precisely because it was too persistent to explain away.

The practical implication is not that everyone should abandon mean-reversion. It is that strategies must be evaluated not just on their average return but on their return distribution shape and their exposure to tail events. A strategy that makes money 90% of the time but blows up on the other 10% is not robust. It is a time bomb with a pleasant-looking equity curve.

For anyone running systematic portfolios, the takeaway is straightforward: ensure your core allocation has positive-skew exposure. Trend-following, volatility targeting, and momentum overlays all share this property. Mean-reversion can complement these as a diversifier, but it should never be the structural backbone of a portfolio that needs to survive across regimes.

References:

Carhart, M. (1997) 'On persistence in mutual fund performance', The Journal of Finance, 52(1), pp. 57-82.

Chen, Y. (2026) 'Be Water: An Evolutionary Proof for Trend-Following', arXiv preprint, arXiv:2603.29593. Available at: https://arxiv.org/abs/2603.29593 (Accessed: 28 June 2026).

Jegadeesh, N. and Titman, S. (1993) 'Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency', The Journal of Finance, 48(1), pp. 65-91.

Moskowitz, T., Ooi, Y. and Pedersen, L. (2012) 'Time series momentum', Journal of Financial Economics, 104(2), pp. 228-250.