For many years the Stocks, Bonds, Bills & Inflation yearbook has served as the primary source for calibrating historical asset returns. However, uneasiness has grown about its depiction of corporate bond returns prior to the second World War. I document problems with the source data used in the SBBI and replace its flawed dataset with new observations of bond prices from 1926 to 1946 for a sample of several hundred large bonds listed on the NYSE and rated investment-grade. I find that the SBBI overstates corporate bond returns in the 1930s and accordingly, gives an unreliable estimate of the premium received for owning investment grade corporate bonds rather than government bonds during the prewar years. To extend the analysis I collected additional bond price data from 1946 to 1974 and find that the SBBI also overstates corporate bond returns in the 1960s. The problem again stems from a reliance on flawed yield series in place of observing bond prices. I combine the new data with existing data to examine the corporate bond premium from 1909 through 2019. Using ten-year rolling returns, over the past century I find the average premium earned on long maturity corporate bonds to be small, about 15 basis points annualized. For many of the ten-year rolls, the premium was instead a deficit: a bond investor would have done better owning only long government bonds. The small and fitful premium contrasts with the yield spread on investment-grade bonds, which was always positive and substantial throughout the period. Because the premium has been much more variable, the relative size of the yield spread does not seem to be predictive of whether a premium will subsequently be earned and how much. Results are interpreted in terms of the importance of regime change in financial history: sometimes corporate bonds outperform government bonds, sometimes they do not, just as sometimes stocks outperform bonds, and sometimes they do not, contra Siegel (2014). The idea of regime change challenges the notion that a mean computed over a longer rather than a shorter interval contributes any additional predictive power to the study of asset returns over human horizons.