TL;DR: A Trigr agent can now run 52 strategies, one per market. That does not buy 52 independent bets. Crypto perps share a common driver, and strategies with similar rules trade on the same days. A useful rule of thumb puts N strategies with average pairwise correlation ρ at about N / (1 + (N − 1)ρ) independent bets. At ρ = 0.3, 52 strategies are worth about three. Mixing styles, timeframes and markets raises that number more than adding strategies does.
Why does correlation decide how much you diversify?
Diversification works when the parts of a portfolio have bad periods at different times. The measure of that is correlation: a number from −1 to +1 that says how closely two return series move together. At +1 they move in lockstep. At 0 they are unrelated. At −1 one gains when the other loses.
This idea sits at the core of modern portfolio theory. The volatility of a portfolio falls as you add holdings, but only down to a floor set by how correlated those holdings are. Past that floor, more holdings add work and fees without lowering risk.
Strategy correlation vs asset correlation
Most traders think about correlation between coins. For an agent, the number that matters is the correlation between strategy returns, and the two can differ a lot.
A strategy's return on a given day depends on whether it holds a position and which way. Some examples:
- Same coin, different styles. A trend follower and a mean-reversion strategy on ETH often hold opposite views. Their returns can be close to uncorrelated even though both trade one asset.
- Different coins, same style. A long-only trend strategy on SOL and one on AVAX enter on rallies and exit on breakdowns. SOL and AVAX move together, so the two strategies win and lose on the same days.
- Time in market. A strategy that is flat 80% of the time has no return on those days. Two such strategies can look uncorrelated in calm periods and then lose together on the few days both are in trades.
The last point is the trap. A low average correlation can hide high correlation on exactly the days that set your drawdown.
Why do trend systems on altcoins move together?
Altcoin perps carry a large share of Bitcoin's moves. When BTC breaks down, most altcoins break down with it, often harder. Trend rules then fire across the board:
- Moving-average crossovers and breakout levels on many coins trip within a few bars of each other.
- Long-only systems all exit, or stop out, in the same session.
- Two-way systems all flip short together, which works well until a sharp reversal stops them all out together.
So a 40-strategy book of trend followers on 40 altcoins can act like a few large bets on the direction of crypto. Our BTC trend-following guide shows how one of those systems behaves on its own. Forty of them on correlated coins behave much the same way, just with more trades.
How many independent bets do you really have?
A plain rule of thumb answers this. Take N strategies with equal weights and similar volatility, and let ρ be the average correlation between every pair. The effective number of independent bets is roughly:
N_eff ≈ N / (1 + (N − 1) × ρ)
When ρ is 0, N_eff equals N. When ρ is 1, N_eff is 1, however many strategies you hold. As N grows, N_eff approaches 1 / ρ and stops there.
| Strategies (N) | ρ = 0.1 | ρ = 0.3 | ρ = 0.5 |
|---|---|---|---|
| 6 | 4.0 | 2.4 | 1.7 |
| 20 | 6.9 | 3.0 | 1.9 |
| 52 | 8.5 | 3.2 | 2.0 |
| Ceiling (1 / ρ) | 10 | 3.3 | 2 |
Going from 6 to 52 strategies at ρ = 0.3 moves you from about 2.4 to about 3.2 effective bets. Bringing ρ down from 0.3 to 0.1 at 52 strategies moves you from 3.2 to 8.5. Lower correlation does far more than a longer list.
The same math shows up in volatility. With equal weights, equal volatility σ and average correlation ρ, portfolio volatility is about σ × √(ρ + (1 − ρ) / N). At ρ = 0.3, six strategies run at about 65% of one strategy's volatility. Fifty-two run at about 56%. The floor, with infinitely many, is about 55%.
A worked example
Compare two books.
- Book A: 40 trend-following strategies on 40 altcoins, with an average pairwise correlation of 0.5. N_eff = 40 / (1 + 39 × 0.5) = 40 / 20.5 ≈ 1.95.
- Book B: 12 strategies split across trend, breakout, mean reversion and two-way styles, on mixed timeframes and a few TradFi markets, with an average pairwise correlation of 0.15. N_eff = 12 / (1 + 11 × 0.15) = 12 / 2.65 ≈ 4.5.
Book B holds 28 fewer strategies and more than twice the independent bets. With fewer strategies, each gets a larger allocation, so each entry clears Hyperliquid's $11 minimum order more easily on a small account.
The formula is a rough guide. Real books have unequal weights, unequal volatility and correlations that change over time. Use it to compare setups, and read the answer as an order of magnitude.
How can you diversify on Trigr?
Each of these levers lowers correlation between strategies.
- Mix styles. Combine trend, breakout, mean reversion and two-way strategies. Our mean-reversion guide covers a style that often wins when trend systems chop.
- Mix timeframes. A 4-hour strategy and a daily strategy on the same coin enter for different reasons and hold for different lengths. The multi-timeframe strategy guide shows how to combine them in one strategy, too.
- Add TradFi markets. Gold, oil, the S&P 500 and single stocks trade as perps on Hyperliquid through HIP-3, and their drivers differ from crypto's. The HIP-3 guide explains the markets and how far Trigr supports them today.
- Start from a Low Correlation preset. On the ready-made agents page, the Low Correlation presets pick members so that no two have a monthly-return correlation above 0.4. Monthly correlation can understate how strategies move together within a crash week, so treat that cap as a starting filter. Our presets guide covers the other families.
- Use allocation modes with care. The builder offers Equal, By Sharpe, Inverse vol and Custom allocations. Inverse vol gives smaller allocations to more volatile strategies, which keeps one volatile coin from dominating the book. It does not reduce correlation.
How do you check diversification in the combined backtest?
The builder's combined backtest gives you two useful readings.
Drawdown. Compare the combined maximum drawdown with each member's own drawdown on its strategy page. If the combined figure sits close to the typical member's, the strategies lost together and the book is less diversified than its length suggests. If it is much smaller, losses were spread out in time. Check the "Data through" date too: the preview may be up to one day stale.
Skipped entries. When allocations total more than 100%, the preview runs one shared-margin simulation of the whole book. Entries that do not fit the free margin are skipped, and the preview counts them and names the strategies that skipped most. Skips that bunch on the same dates show signals firing together, which is correlation in plain sight. Our post on position size vs allocation explains the first-come rule behind those skips.
Both readings come from history. A combined backtest of strategies chosen because they looked good together will look better than it trades forward. Run the book on a paper agent and watch how many positions open on the same day.
What diversification cannot do
Diversification lowers the risk that one strategy's bad patch sinks the book. It does not remove market risk. In a sharp crypto sell-off, correlations between strategies tend to rise, and a long list of strategies can lose together. Size the agent for that day: assume most crypto positions open at once lose together, and keep leverage modest. Our multi-strategy portfolio guide covers sizing and shared margin in more detail.