Compare allocations using the same strategies
Equal weighting gives each strategy the same share of capital. Risk parity chooses weights with the strategies' contributions to risk in mind. The question for machine learning is whether it can improve the weighting decision relative to these rules while using the same set of investments.
The comparison uses daily returns for ten strategies from February 2022 through November 2025 and sets weights monthly for each approach. Holding the strategies and period fixed helps separate the difference due to allocation from a difference due to one portfolio having access to better strategies.
What drove the risk-adjusted difference?
The machine learning allocation recorded the highest Sharpe ratio in the sample, with the improvement driven mainly by higher returns. The Sharpe ratio scales return above the risk-free rate by the portfolio's own volatility. The information ratio relative to risk parity was about 0.94. It scales return above that benchmark by the volatility of the difference.
The Sharpe ratio alone cannot show how consistently the portfolio beat risk parity. The two measures describe the portfolio's own risk-adjusted performance and its performance relative to a specific allocation rule. The comparison is evidence about the ten strategies and the period analyzed.
Test the difference in other periods and after costs
An allocation model needs to show how it performs in periods kept apart from fitting and tuning. If it changes weights frequently, turnover and transaction costs can also affect whether a difference in returns is useful in practice.
A comparison with total-return TOPIX and a factor-based analysis can help examine which market risks contributed to the observed difference. The value of the allocation method depends on the conditions under which it can reproduce that difference, as well as its ranking in the original sample.