XIV Bias in Industry Data 1 Bias in
XIV. Bias in Industry Data 1
Bias in Industry Data Almost all mutual fund services list as of a date the record (often over a quarter, a year, a three year period, a five year period, and often a ten year period) of all funds that meet a criteria, (e. g. size) as of a certain date. Common practice to represent the average return for the industry on what a typical investor might earn by averaging these results. Overstatement of what industry did because of ignoring funds an investor might have bought which did so badly they are no longer in existence and thus no longer in the data base. Produces misleading conclusions about types of funds which have characteristics associated with different rates of attrition. (Incobator Funds) 2
Common Stock Funds Trace Period January 1976 -December 1993 322 Funds Survived or Merged Survived Merged (DIM) 67. 0% 33. 0% Survived, Merged, or Policy Change Survived Merged Policy Change 60. 1% 23. 5% 16. 4% 3
How do you estimate bias? Sample – All plain vanilla common stocks funds that had 15 million dollars or more under management at the end of 1975 Used a follow the money approach. Traced all name changes and mergers and policy changes for every fund. If merged or changed policy (stock or bond) found values at date of change. Assumed investor stayed in acquiring fund unless policy changes. If policy change reallocated to other funds in sample. Contacted fund families, master trustee, auditing firm. Traced all but 1. 4
5
SURVIVORSHIP BIAS SAMPLE LENGTH 10 YEARS ORIGINAL SAMPLE 46 FUNDS MERGED 10 FUNDS FAILURE RATE 1 PER YEAR OR 2% AFFECT ON SAMPLE 27 BASIS POINTS PER YEAR IMPLIES RETURN FROM BOND DATA BASE ABOUT 27 BASIS POINTS TOO HIGH 6
Acquired and Acquiring DIM SAMPLE ACQUIRE R 1. Avg. Size (Millions) 52 305 155 Rank (Percentile) 25 43 2. Funds merge into funds which have higher expense ratios than DIM funds and average funds in sample. (All corrected for size) 3. Performance before merger (3 Year α) DIM ACQUIRER -3. 6 +1. 7 After merger remaining fund has lower α (does worse) than average fund in sample 7
Fund Characteristics If don’t correct for bias small funds do better If correct small funds do worse If don’t correct overestimate the performance of higher risk funds (e. g. , maximum capital gain funds) relative to safer funds (e. g. , growth and income funds) Conclusions Whenever you look at a data base or a study think of survivorship bias. (Compustat, Manager Performance) 8
- Slides: 8