Nonparametric Walk-Forward Backtest Protocol

You have daily returns for $N$ candidate trading signals $\{f_i\}_{i=1}^{N}$. For each signal, you form a dollar-neutral portfolio by ranking stocks on the signal and going long the top decile, short the bottom decile. 1. Propose a rolling walk-forward scheme that produces an out-of-sample Sharpe ratio estimate for each signal. Specify how you choose the training window length, the rebalance frequency, and the test window length. 2. With $N$ signals tested, how do you account for multiple testing? Describe a nonparametric procedure to control the false discovery rate (FDR) so that you can identify which signals, if any, have genuinely positive out-of-sample Sharpe ratios. 3. What are the key pitfalls of this approach, and how would you mitigate them in practice?