Standard asset-pricing factors are usually treated as single return series, yet the risks they summarize need not be homogeneous across horizons. I decompose the market, momentum, and value factors into orthogonal spectral components using a downsampled Haar discrete wavelet transform and estimate sparse mimicking portfolios for each band with elastic net and band-specific complexity controls. The results reveal clear heterogeneity across factors. Market risk is broadly distributed across horizons and remains economically trackable even at longer horizons. Momentum is largely a short-horizon object: tracking is strongest at high frequencies and spanability deteriorates quickly as the horizon lengthens. Value looks different from both, with more informative variation shifting toward medium and lower frequencies, but with weaker and less even traded representation. The clearest boundary case is HML D5, roughly 32--64 trading days, for which the full-sample procedure does not recover a stable mimicking portfolio.
Comissão Organizadora
Comissão Científica