Factor turnover measures how frequently the assets associated with a given quantile change, that is, how many trades are required to adjust a portfolio to the sequence of signals. More specifically, it measures the share of assets currently in a factor quantile that was not in that quantile in the last period. The following table is produced by this command:
create_turnover_tear_sheet(alphalens_data)
The share of assets that were to join a quintile-based portfolio is fairly high, suggesting that the trading costs pose a challenge to reaping the benefits from the predictive performance:
|
Mean Turnover |
5D |
10D |
21D |
42D |
|
Quantile 1 |
59% |
83% |
83% |
41% |
|
Quantile 2 |
74% |
80% |
81% |
65% |
|
Quantile 3 |
76% |
80% |
81% |
68% |
|
Quantile 4 |
74% |
81% |
81% |
64% |
|
Quantile 5 |
57% |
81% |
81% |
39% |
An alternative view on factor turnover is the correlation of the asset rank due to the factor over various holding periods, also part of the tear sheet:
|
5D |
10D |
21D |
42D |
|
|
Mean Factor Rank Autocorrelation |
0.711 |
0.452 |
-0.031 |
-0.013 |
Generally, more stability is preferable to keep trading costs manageable.