Factor outperformance for more than two centuries
To be considered relevant, a factor must first and foremost be backed by ample empirical evidence. In the absence of such evidence, academic research on multi-asset factor premiums could suffer from ‘p-hacking’ (or ‘data mining’). Recent research uses new and previously unused deep historical financial data. The results allay any p-hacking concerns.
Speed read
- Potential ‘p-hacking’ is a serious threat to financial research
- Factors should be backed by overwhelming empirical evidence
- New research by Erasmus University and Robeco colleagues looks at over two centuries of financial data
When assessing the evidence on potentially interesting factors, it is of utmost importance for researchers to take the possibility of ‘p-hacking’ into account. P-hacking means that researchers examine thousands of different investment strategies, but only report those that show the best historical performance. The discovery of new historical financial data is important to alleviate these p-hacking concerns. The quest for such historical evidence on factor premiums was actually what recently led Guido Baltussen, Laurens Swinkels and Pim van Vliet to analyze more than two centuries of international market data from multiple historical sources, relating to an array of asset classes.
The three authors looked at six major factor premiums in equity indices, government bonds, currencies, and commodities, using data going back to 1800. Such an extensive sample allowed them to falsify certain factors, to check if previous results were simply driven by ‘p-hacking’. A data set spanning such a long period of time covers the various stages of the business cycle and economic crashes, which makes it possible to study the sensitivity of factor premiums to financial market and macroeconomic conditions.
Persistent factors
The authors found significant, persistent, and robust momentum, value, seasonal and carry premiums within four asset classes. They document the presence of a low-risk effect in equity markets, but not in other markets, which is consistent with the typical explanations given for the low-risk effect. They also found that the momentum, value, seasonal, and carry factor premiums generally work well with each other in the context of portfolio diversification. The authors also show that the time-series trend and cross-sectional momentum are in essence similar factors. Further, they argue that multi-asset factor strategies give rise to statistically highly significant Sharpe ratios in the range of 0.5-1.2, delivering positive returns in almost all 10-year periods since 1800. The authors concluded that the factor premiums remain robust across economic stages
The factor premiums remain robust across economic stages
The authors observed that the factor premiums remain robust across economic stages, such as bull and bear markets, recessions and expansions, and periods of crisis and of growth and prosperity. They also investigate downside crash risk as a possible explanation for these factor premiums but find virtually no evidence for this.
Multi-asset factor investing
Overall, we believe the authors present convincing empirical evidence that most factor premiums are present in all major asset classes, are economically robust, and persist over time. They are highly unlikely to be the result of data mining, supporting the assumption that they will continue to exist in the future.
Read the related research paper, ‘Global Factor Premiums’, on SSRN.
