A portfolio of 100 strategies sounds safer than a portfolio of 10. It is safer only if those strategies do different things. If twenty of them win and lose on the same days, they are one position with twenty names, and the risk budget that treats them as twenty separate ideas is wrong by a factor of twenty.
This is one of the most practical questions in systematic trading, and it is rarely answered with numbers. NOVAQ has measured it twice, with different methods, and the answers agree.
Measuring what a book really does
In June 2026, daily results of 36 atomic strategies were correlated and clustered hierarchically on two levels. The first level joined near-duplicates into 18 micro-streams. The second joined related streams into 12 macro bets. The largest merger came in short-term momentum, where about 20 scalping variants formed a single stream.
In October 2026, a different measure was applied to 55 strategy books. The effective number of exposures is calculated from the correlation matrix and answers a simple question: how many uncorrelated positions would behave like this set? The answer was about 12. Two methods, four months apart and on different books, landed in the same place. The figure measures diversification and is not a count of proven edges.
From strategies to independent bets (June 2026)
What diversification looks like in the data
Mean daily correlation between books was about 0.13. The more valuable number sits underneath it. Correlations often rise exactly when a portfolio needs them to fall, as different strategies get caught in the same market shock. On the worst 10% of days for this book, mean correlation between its components stayed close to zero.
Daily correlation between strategy families
The portfolio result
53 rule-based books at equal weight, after costs, from 2020 to October 2026: Sharpe about 2.15, return to drawdown about 23, all 7 calendar years positive (2026 partial) and 57 of 82 months positive. Of the 55 books in the wider set, 52 finished net positive.
The main families each contributed in all 7 calendar years, with different profiles. Mean reversion reached a Sharpe of 2.25, opening-range and breakout strategies a return to drawdown of 19.2, while momentum and trend added further distinct exposures. Combining families with non-overlapping drawdowns lifted return to drawdown to 25.5, above the 20.0 of the best single family. That difference is what diversification is worth when it is real.
53 rule-based books, cumulative result and drawdown (R)
How it changes the design
These measurements shape the way NOVAQ is built. Risk budgets are set per exposure and not per strategy. Diversification is tested on shared calendar days, so simultaneous losses are counted as simultaneous. A component with a weak standalone score can stay in the book if it feeds or offsets another. New strategies are judged by what they add to the whole, and a strong result that repeats an existing exposure adds less than its numbers suggest.
The analysis depends on the rest of the platform. Correlations between strategies mean little if their inputs are faulty, which is why this work rests on the clean data corpus and the point-in-time review of the book.
Limits
Most individual strategies were tested with separate in-sample and out-of-sample periods, while this figure aggregates the historical results of the selected strategies. About three quarters of the return is long-side in a rising market, and the result comes from NQ. 2025 was the weakest full year. Days without a trade count as zero, which lowers measured correlation. R uses one unit per book, so these are not account results.