Thread by @GestaltU

TWEETSTORM on 200 Years of Global Premiums

1/ @Paradoxinvestor and his team at @Robeco spent months constructing a deep historical dataset stretching back to 1800 consisting of stock indexes, 1-year government bonds, commodities and currencies.

Summary:
investresolve.com/blog/over-two-…

2/ They used this novel dataset to conduct long-term out-of-sample analyses on some of the most pervasive, persistent, economically significant and investable style premia: trend, momentum, value, carry, seasonality, and Betting-against-Beta (BAB).

3/ In all they tested 6 major premia across 4 asset classes for a total of 24 tests over two centuries. Their results unambiguously “reveal strong global factor premiums that present a challenge to asset pricing theories.”

4/ This paper focused on factors at the asset class level and produced three novel contributions:

• Replicated seminal factor studies dealing with multi-asset premia using a uniform testing methodology over the original period 1981 – 2011.

5/cont…
• Applied the same uniform testing methodologies to extend their tests back to 1800

• Examined the relationships between style premia returns and common macroeconomic and financial risk factors to determine if any of the premia have compelling risk based explanations.

6/ Sharpe ratios averaged 0.41 with uniform specification in the replication study. Half of the premiums were significant at the 5% threshold, while 1/3 were significant at the 1% threshold. Multi-asset-class versions produced Sharpe ratios between 0.39 (BAB) and 1.15 (Carry).

7/ The paper applied more rigorous tests of significance to account for the p-hacking effect, where a researcher will surface what appears to be significant results simply from running many similar tests.

8/ Specifically, they provided Bayesian p-values for all tests using a 4:1 prior odds ratio, consistent with a threshold that expert Cam Harvey classifies as “perhaps” sufficient to address p-hacking concerns.

9/ The figures below illustrates the Sharpe ratios from the original papers (left) and from the uniform replication tests (right). Numbers above the bars represent more conservative Bayersian p-values (lower values  small chance that the results were due to luck).

10/ Trend and carry are dominant factors in both the original and the replication samples, with statistical significance in excess of conservative thresholds in all asset categories. The uniform multi-asset versions of trend and carry both produced t-stats greater than 6.

11/ Value, seasonality and BAB showed only marginally significant effects in the replicating sample, even against the more tolerant frequentist thresholds (traditional p-values of 0.06, 0.05 and 0.4 respectively). The effects fell well below Bayesian thresholds.

12/ In the almost 200 years out-of-sample the authors found an average Sharpe ratio across factors of 0.41, exactly consistent with what they found in the replicating sample from 1980-2012.

13/ The larger sample validated results with much higher levels of significance; 19 of 24 combinations of factors and asset categories produced t-values above 3. Trend, carry and seasonality showed the greatest promise per Panel C below.

14/ Return seasonality in bonds and FX was especially strong in the new data. Momentum and value were also significant but FX value and commodity momentum were notable exceptions. BAB failed tests of significance in every category except equities.

15/ As expected, diversification across asset categories produced highly compelling results for multi-asset-class strategies across all factors. Multi-asset trend, carry and seasonality all exhibited t-stats greater than 15 (‼) exceeding even conservative statistical thresholds.

16/ Factor specification can play a material role in results, even over long sample horizons. The authors performed tests on a variety of methodological variations to test for robustness. Overall the authors concluded that results were robust to alternative specifications.

17/ They removed the liquidity screen (increased Sharpe ratios); formed equal weight tertile portfolios (no change); eliminated volatility scaling (over-weighted high volatility instruments and lowered aggregate Sharpe ratios); …

18/ …lagged signals by one month (small decay in most strategies but completely eliminated seasonality effect by construction; rebalanced quarterly (small decay in most strategies except value); and trimmed extreme returns (slight increase in performance).

19/ The authors examined several macroeconomic and financial risk-based explanations for each premium but found no meaningful or statistically significant relationships despite 43 bears markets and 218 downside market states in their sample.

20/ The authors conclude “In summary, our tests reveal very limited evidence of a link between macroeconomic risk and global return factors.”

21/ Finally, given that the premia have produced strong returns with low sensitivity to markets and traditional risk factors, allocations to risk factors have the potential to substantially expand the efficient frontier, and add value to most portfolios.