This paper studies the time series effect of changes in liquidity on optimal portfolio allocations. Using a nonparametric approach, we are able to handle models that are analytically intractable. Specifically, we directly estimate optimal portfolio weights for a CRRA investor as functions of liquidity. Liquidity is measured by turnover, dollar volume, or price impact. We consider three different investment horizons: daily, weekly, and monthly. Using a sample of NYSE stocks from 1963-2000, we document a very interesting temporal dimension to the effects of changes in liquidity: whereas optimal weights are strongly increasing functions of liquidity at the very short daily and weekly horizons, they become decreasing functions of liquidity at longer monthly horizons. Overall, the dependence of optimal weights on liquidity is most noticeable for small stocks at short investment horizons. Finally, the optimal conditional portfolio weights documented in this paper are never negative, which may help explain the low level of short selling observed in the US stock market.