We recently read a research paper regarding a somewhat surprising result – that for CTAs, it did not pay to be allocated to diversified managers. So, as NilssonHedge has a large dataset of managers, we decided to do a similar exercise, but with different filters to test the robustness of the results. We came to a different conclusion. We will write up something more formal at a later stage, with additional checks and balances. Below are preliminary results.

Here we would like to point out that this is not a realistic portfolio as it switches between managers at a furiously high pace. No organization would be able to perform due diligence at that speed, no broker would be able to open accounts as needed. And most managers would probably shun a strategy like this. However, as a theoretical exercise, it is interesting.

- We got all the managers that we have classified as Trend Followers (here we got a difference as the paper used cluster-based analysis to find managers). The highest manager count was 243 and the lowest was 84 early in the sample.
- We calculated the rolling 24-month correlations for each manager against the average performance of all the managers.
- Based on all the correlations, we divided the sample into quartiles (the paper divided the returns into quintiles).
- The quartiles were used to calculate four variable portfolios of managers, designed to capture managers that had varying degrees of correlation to the benchmark. We lagged the correlation estimate one month, to ensure that we did not have a look-ahead bias.
- Besides, we risk-adjust the portfolio to create portfolios that fully utilized the diversification effect.

We disregarded survivorship (as our historical data does not extend far enough), avoided any failure penalty (the paper used a negative 50% penalty for CTAs that stopped reporting). No assets under management filters were applied.

In the database, we have 464 managers in this classification, but no more than 243 are alive at the same time. There is definitively some turn-over over the last 20 years. Our sample is far from complete in the early period but looks more reasonable in the latter half. The tail at the end is natural as not all managers have reported final numbers.

The reporting bias in the database may favor managers with shorter track-records. However, we would suspect that diversifying managers would have slightly easier to raise assets, at least when trend following strategies are having unappealing performance. We required that a manager had a 24-month correlation history before being considered for the portfolio.

Correlation clusters over the last 20 years show variability, but nothing out of the ordinary. We note that the median Trend Following manager correlates 0.8 to the group. The groups are dynamic and the constituents change over time, based on the correlation behavior.

The highest group seems to represent a more style pure manager. Aspect Diversified, EMC Classic, Estlander Freedom and Man AHL are frequently found in this group. These are all fairly well-known representatives of Trend Following strategies.

After having clustered each manager into a temporary correlation ranking, we create portfolios consisting of the 1st to the 4th correlation quartiles.

Here we roughly consistent results with the paper, the least diversified quartile produces better returns, but the diversification benefits provide rebalancing gains that offset the higher average returns. The 4th quartile (the most correlated managers have the highest average annual return) but the Full portfolio consisting of all managers compounds more efficiently due to a marginally lower return, but much lower risk. The absolute return of the portfolios are shown in the table below. These are consistent with the idea that diversification does not pay for CTAs.

Full Sample | 1st Quartile | 2nd Quartile | 3rd Quartile | 4th Quartile | |

Return | 7.3% | 6.0% | 6.8% | 6.4% | 7.4% |

Risk | 11.4% | 7.6% | 12.9% | 14.5% | 14.9% |

Sharpe | 0.53 | 0.62 | 0.42 | 0.35 | 0.41 |

As the portfolios are not risk-adjusted, we adjust the portfolio to create portfolios with the same target volatility (we subtract the prevailing risk-free rate of return, adjust for volatility and add back the risk-free rate of return).

Full Sample | 1st Quartile | 2nd Quartile | 3rd Quartile | 4th Quartile | |

Av Ret | 5.3% | 6.2% | 4.2% | 3.5% | 4.1% |

Risk | 10.0% | 10.0% | 10.0% | 10.0% | 10.0% |

Sharpe | 0.53 | 0.62 | 0.42 | 0.35 | 0.41 |

After risk adjusting the portfolio, we observe how much potential return that the additional Sharpe is potentially worth. More interesting, we note that the portfolio consisting of all managers is the runner up.

Ranking the portfolio on a 12-month rolling basis, to determine which ones have the greatest odds of having the best performance, the Full Sample portfolio is the one with the highest-ranked performance, i.e the portfolio that is most likely to win over the other portfolios.

RankW | 481 | 577 | 642 | 753 | 652 |

Rank | Full Sample | 1st Quartile | 2nd Quartile | 3rd Quartile | 4th Quartile |

1 | 27 | 79 | 27 | 23 | 51 |

2 | 105 | 15 | 41 | 20 | 26 |

3 | 56 | 34 | 52 | 42 | 23 |

4 | 19 | 29 | 58 | 46 | 55 |

5 | 0 | 50 | 29 | 76 | 52 |

In the end, the Portfolio with the best Sharpe ratio looks to be the most diversified one. The portfolio with the highest odds of outperforming the other portfolio on a 12-month basis is the full portfolio. The style purest portfolios end up in the middle, muddling through, but not more.

Based on preliminary results, diversification still pays, but we may need to adjust for survivorship and reporting biases. However, we need to be able to use notional funding and to adjust managers up and down based on the risk and correlation that they exhibit.

These results are preliminary and once we have time, we will add a few more filters, expand the selection and perhaps try a clustering algorithm to classify managers.

You must log in to post a comment.