Trade Like A Chimp! Unleash Your Inner Primate

ELI5/TLDR

Andreas Clenow runs random monkey portfolios against the S&P 500 and most of them win. The reason is not that monkeys are smart, it is that the S&P 500 is a market-cap-weighted index where ten giant stocks swamp the bottom three hundred. Once you stop letting the index pick your weights, almost anything works — random stocks, random weights, random number of stocks. The lesson is that beating the index is trivially easy, the real benchmark for any quant strategy is whether it beats the chimps.

The Full Story

The setup: indices are not investment strategies

Clenow opens by reminding the Quantopian crowd that the S&P 500 has spent long stretches doing nothing. Ten flat years with a fifty percent drawdown in the middle, fifteen years compounding at five percent. Mutual funds fare worse — eighty to eighty-five percent of them fail to beat the benchmark over any given three, five, or ten year window. He stresses that this is the fault of the model, not the managers.

The set-up is a Burton Malkiel reference, the blindfolded monkey from A Random Walk Down Wall Street. Clenow takes it literally.

Empirical observation does prove the point that monkeys know how to select stocks.

He cites a 1993 Swedish stock-picking competition where an actual chimpanzee named Ola beat the country’s top analysts. The chimp died in Thailand a decade later, which is its own story.

The monkey simulations

The actual experiment is three escalating tests, all rebalanced monthly, all drawing only from contemporaneous S&P 500 members (he is careful to note he uses point-in-time membership — Yahoo Finance will not do).

Random stocks, equal weight. Fifty randomly selected names, equal dollars each. Most simulations beat the S&P 500 total return.
Random stocks, random weights. Same fifty names but with no constraints on allocation — one stock could be ninety-nine percent of the portfolio. Bigger drawdowns, wider spread, still beats the index on average.
Random number of stocks (5 to 300), random weights. This should not work. It does.

His point lands hard. The S&P 500 is a market-cap weighted index. The top ten names have roughly the same weight as the bottom three hundred. So it is not really a 500-stock index, it is a handful of mega-caps with three hundred decorative names underneath. And since stocks enter the index because they have already run up, the index is, in his words, “a kind of momentum index with a very strange weighting scheme.”

Equal-weight and volatility-parity: the cleaner cuts

Strip out the market-cap weighting and the random portfolios outperform. The official S&P 500 Equal Weight index (yes, it exists, you can buy the ETF) has always beaten the cap-weighted version over long periods. Equal-weight is the boring version of the monkey trick.

Clenow’s preferred sizing is inverse volatility — what the quant crowd calls volatility parity. The logic is plain. If you put a million dollars in a sleepy utility and a million dollars in a biotech, the biotech runs the portfolio. So you scale position size by inverse ATR (he uses 20-day ATR) so each stock contributes roughly the same risk. He notes in passing that “risk parity” is the marketing term; “volatility parity” is what it actually is.

Random fifty stocks at volatility parity is the best-performing random strategy in his deck.

The “be the better primate” strategy

If you cannot tell your clients you are running a random number generator, you need a portfolio model that beats the monkeys. The four pieces:

Universe. S&P 500, current members, with the graveyard accounted for.
Screen. Positive momentum only. No buying falling knives.
Rank. His “cleaner momentum” — annualized exponential regression slope times the R-squared of the fit. Exponential because percentage moves are comparable across price levels. Annualized so the number is readable. Multiplied by R-squared so a stock that jumped a hundred percent in a day on a takeover gets penalized for being a poor fit to a smooth trend. He wants the firm slow ascent, not the spike.
Size and rebalance. Volatility-parity sizing. Monthly rebalance. Buy from the top of the rank until cash runs out. Floor the slope at 30 — if the cleanest stock in the universe scores below that, hold cash. In bear markets the model can sit largely in cash, which he says is a feature, not a bug.

The model beats the S&P 500 total return. More importantly — and this is the actual benchmark of the talk — it sits in the top bucket of the random monkey simulations. Some monkeys still beat it. He is honest about that.

If you end up in the middle of the road in the middle of the monkey strategies, you really didn’t accomplish that much with the rules.

Why this matters as a methodology, not a strategy

The deeper claim is the random-portfolio benchmark itself. Most quant decks compare a strategy to the index and call it a win. Clenow’s point is that the index is a low bar — almost any non-cap-weighted portfolio clears it. The honest test is whether your rules add value above a random portfolio drawn from the same universe. Every additional rule has to pay for itself in lift over the random baseline. Otherwise you are just paying yourself to complicate a coin flip.

He closes with two Q&A asides worth keeping. On shorting: most people should not. Trend-following on the short side tends to break even at best; it earns its place in a diversified book only because it improves skew. On combining strategies for clients: institutional allocators want clean building blocks, not your pre-mixed multi-strategy fund — they want to do the combining themselves.

Key Takeaways

The S&P 500’s flaw is cap-weighting. Top 10 names ≈ bottom 300 by weight. Any other weighting scheme — including random — tends to beat it.
Three random portfolios, all rebalanced monthly, all from point-in-time S&P 500 members, all beating the index: equal weight, random weight, random count + random weight.
The benchmark for a quant strategy should be a random portfolio from the same universe, not the index.
Clenow’s momentum signal: annualized exponential regression slope × R-squared. Slope in percent (so it cross-compares stocks), R-squared as a smoothness penalty.
Sizing rule: inverse 20-day ATR. Each name contributes equal volatility to the portfolio.
Filters: must be current index member, positive slope, slope ≥ 30 annualized, top 20% of the rank. Otherwise hold cash.
Bear-market behaviour: the floor on momentum means the strategy goes to cash, which is what cuts drawdown.
On shorting: trend-following short equities is roughly break-even over long horizons. Include it only for skew, if at all.
On packaging: combining strategies improves Sharpe but hurts the pitch — allocators want single-factor building blocks.

Claude’s Take

Most of the value of this talk is in the framing, not the model. Cleaner momentum is fine, well-trodden by 2017 — the slope × R-squared idea was already in Robert Carver’s territory and adjacent literature. The 20-day ATR volatility parity is textbook. None of the individual components are novel.

What is genuinely good is the random-portfolio benchmark. It is a clean, embarrassing test that most factor strategies probably fail. If your “smart beta” product cannot reliably finish above the median monkey drawn from the same universe, you are selling weighting-scheme alpha and calling it skill. That is a real contribution, and the visual of the chimp lines stacked above the S&P 500 white line does more work than most academic papers on the equal-weight premium.

The “trade like a chimp” packaging is exactly what it looks like — a Quantopian-conference hook for what is otherwise a standard systematic-momentum talk. He admits as much by the time he gets to slide ten. The hedge fund cross-promotion at the end (his books, his ECTA business, his allocation seat) is also doing a job. None of that takes away from the methodology, but it is worth noticing the structure: monkeys to get you in the room, a fairly conventional momentum model in the middle, books on the way out.

The cash-in-bear-markets feature deserves more scrutiny than he gives it. Half the outperformance versus the index is the model sitting out 2008. Whether that counts as “the strategy works” or “the floor at 30 happened to dodge one drawdown” depends on how many bear markets you have in your test window. He has one big one.

Seven out of ten. Strong on benchmarking philosophy, modest on the underlying model, well-delivered.