Portfolio Construction for Many Sparse Trading Strategies

Summary

The post considers how to allocate capital across more than fifty backtested strategies that trade ETFs, mostly equities, with daily signals and next-day open execution. The strategies are mainly short-horizon mean-reversion systems, with different trading frequencies and sparse return series. A simple equal-weight combination reportedly produces an unexpectedly strong equity curve, prompting concern that apparent diversification may be misleading.

The author already knows conventional portfolio optimization and asks whether methods such as mean-variance, decorrelation, concentration, Sharpe, or drawdown objectives make sense for these strategy returns. The document offers no proposed allocation method, diagnostics, numerical evidence, or response, so it is useful chiefly for framing the portfolio-construction problem. Its caveat is central: the strategy set is backtested and only naively validated, and sparse observations can make estimated returns and relationships uncertain. The strong equal-weight result is an observation reported by the author, not evidence that optimized weights or future performance will be reliable.

Key ideas

• The strategies differ in instruments, trade frequency, and return sparsity, complicating allocation across them.
• An equal-weight portfolio is reported to have strong backtest statistics, but the author doubts their reliability.
• The author raises standard optimization objectives while questioning their suitability for this strategy set.
• The post provides no recommended optimizer, robustness analysis, or out-of-sample evidence.
• Naive validation and sparse returns make estimates of strategy performance and dependence uncertain.

Tags

# Combining many trading strategies in an efficient


# Combining many trading strategies in an efficient












I have a lot (>50) of back tested (and naively "validated") trading strategies. They trade different ETFs, mostly equities, but also others (like GLD, USO, ...). These are all strategies developed on daily time frame (daily bars) with entries/exit on open on next trading day. Mostly duration of the trades is a couple of days (mean reversion strategies). They vary in trading frequency. Some trade 30 times per year, some only 5 times per year (so return vector of a strategy is "sparse").

Their combined equity curve and statistics with 1/N allocation is impressive (I have hard time to believe this naive "diversification" of far from perfect trading strategies works so well).

The problem I have is that I would like to use capital more efficiently by having different allocation between strategies than 1/N. I know how to do portfolio optimization and treat each strategy as a return vector and combine them under different objectives (mean variance, max de-correlation, min concentration, max Sharpe, min draw down, ...) and constraints. But I am not sure the "profile" of these strategies lends themselves to this brute force optimization.

What approach would you recommend for combining these kind of strategies in a somewhat "optimal" portfolio? Is there an optimization approach that makes sense here? I would be grateful to be pointed to resources that deal with this kind of strategy portfolio construction.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.