Investing
A practical guide to reading investment research critically, separating robust findings from claims weakened by data mining, implementation costs or changing markets.
The authors warn that if a researcher runs many tests after looking at the data and reports only the ones that came out significant, the significance figure stops meaning what it claims. The 'discovery' can be a random pattern that happened to stand out among many tries, not a real relationship.
Deciding what to test before examining the data, and demanding a higher bar of statistical significance, protects against fishing for results. A claim made after seeing the data deserves much less trust than one predicted in advance.
In the authors' account, a strategy that ignores trading costs, the cost of betting against stocks, and the difficulty of trading thinly traded assets can show gains that vanish in practice. Results that leave out real-world frictions usually can't be repeated with real money.
A model given too many variables, and tuned too hard to past data, starts matching accidental coincidences rather than real effects. It looks impressive on the data it was built from and fails on anything new.
The authors' remedy for chance findings is to split the data. Build the idea on one part, then test it on a part that was never used to build it. A real effect should still show up in the untouched data; a fluke usually won't.
If a study of companies or funds leaves out the ones that failed or shut down, the remaining sample is made up of winners. Average performance then looks better than anyone investing at the time could actually have earned.
The authors argue that an apparent outperformance may simply be pay for carrying known kinds of risk. Judging a result requires adjusting for those risks using current models; otherwise ordinary risk-taking gets mistaken for skill.
The authors frame low-cost index investing as a benchmark that improves active management rather than simply competing with it. Once a cheap market return is available, active managers must justify their fees against it.
The authors treat guarding against rare, severe market crashes as part of deciding how money is split across assets, not as a separate add-on product. How much crash risk a portfolio can bear is built into its overall mix.
Investors face a constant flood of tips and research, and the sheer volume lets good ideas get lost while bad ones spread. The book's answer is a repeatable checklist for judging whether a claim's method is sound before acting on it.