Estimation and inference in low frequency factor model regressions with overlapping observations

A low frequency factor model regression uses changes or returns computed at a lower frequency than data available. Using overlapping observations to estimate low frequency factor model regressions results in more efficient estimates of OLS coefficients and standard errors, relative to using independent observations or high frequency estimates. I derive the relevant inference and propose a new method to correct for the induced autocorrelation. I present a series of simulations and empirical examples to support the theoretical results. In tests of asset pricing models, using overlapping observations results in lower pricing errors, compared to existing alternatives.