Reliable projections of future wildfire risk are important for planning and adaptation decisions.  Statistically downscaled and bias-corrected climate model ensemble products are routinely used to analyze regional physical wildfire risk, but more work is needed to compare ensemble hindcasts with historical trends and variability.  Here we evaluate physical fire risk over the western United States using the Fire Weather Index (FWI) Risk from statistically downscaled CMIP5 results (MACA) and the underlying observational target data set, gridMET. We analyze multidecadal trends and interannual variability in seasonal average FWI for the historical period and future projections for RC4.5 and RCP8.5 scenarios, and we compare CMIP5 ensemble results with a simple time series model that generates future FWI projections based on bootstrapping observed historical trends and variability. Our findings indicate CMIP5 models typically underestimate historical FWI trends, which may in turn underestimate future increases in fire risk, compared with the simple time series model that projects historical trends and variability into the future. Results can potentially be useful for informing broader multi-sector applications.