Daily Speculations The Web Site of Victor Niederhoffer and Laurel Kenner

Dr. Alex Castaldo

11/07/2005
Hsu & Kuan: Reexamining the profitability of technical analysis with data
snooping checks
Journal of Financial Econometrics, 2005, Vol 3 #4, p 606-628

They considered a "universe" of 39,832 trading rules applied to four stock indexes (DJIA, S&P 500, Nasdaq Comp, and Russell 2000) for the period 1990-2003.

Among the rules tested are 497 Alexander Filter Rules, 2049 Moving Average rules (using moving average periods of 2,5,10, ..., 250 days in both Simple and Double MA (crossover) formula, with or without a protection band), 1220 support/resistance rules, 2040 channel breakout rules, etc. (the full list on page 611). You get the idea. Some rules only use prices and some use volume as well.

With so many rules it is guaranteed that something will appear to work. The authors are aware of this problem and apply Hal White's methodology to compute a significance level that takes the multiple comparisons into account.

Briefly, White's method (the so called "reality check") is a Monte Carlo simulation method. The trading rules are applied both to the real price data and to 1000 artificial data series generated ("bootstrapped") from random reshuffling of the data. The performance of the best trading rule on the real data is compared to the bootstrapped distribution of best rule performance. For example if the best trading rule on real data outperforms 950 out of 1000 artificial runs we could conclude that "it works" at the 5% level.

Because White's method has been criticized by Hansen, they also apply Hansen's proposed method, SPA as an alternative. The difference between the two methods' results are fairly small.

The results of White's method are as follows:

```              Prob.
DJIA          0.39
S&P500        0.22
NazComp      <0.01
R2000        <0.01```

The trading rules do not work for DJIA and S&P 500 but do work for Nasdaq Comp and Russell 2000. The authors attribute this to the fact that the latter two are "relatively young markets" whatever that means (!!!).

The best rule is a SIMPLE TWO DAY MOVING AVERAGE WITH (for Naz) OR WITHOUT (for Russell) AN 0.1% GUARD BAND. This produces annual returns of 38.19% for Naz Comp and 46.99% for the Russell 2000.

CRITIQUE

The paper is not very good from an academic or practical point of view.

From an academic point of view it is not very original, the key idea is lifted from White. It is also, in some ways, a remake of the Brock, Lakonishok and LeBaron paper of 1992 but with more recent data and more trading rules. On the plus side the authors did do an impressive amount of programming and calculation.

From a practical point of view: The equity curve for Naz shows that the rule worked terrifically from 1990 to 1999, after which it fluctuated erratically and then started to decline (after transactions costs). For Russell the returns were spectacular from 1997 to 2000, a peak was reached in 2001 and again the curve headed down. "It used to work" may be the best conclusion.

Prof. Pennington Remarks:

The rule that worked for the Nasdaq composite and Russell 2000 is based on very short term momentum. Ask any mutual fund timer and he'll tell you that what they're seeing is stale pricing. Stocks go up, but many components of the Russell 2000 don't even trade, and so they don't appear to go up until the next day, or maybe even a few days later. And isn't it interesting that the rules only work for the Russell 2000 and the Nasdaq Composite!

Alessandro Castaldo, CFA, is a researcher and trader for Manchester Trading. Dr. Castaldo wrote his PhD dissertation on stock market volatility at the City University of New York, and taught courses in finance and options to undergraduates at Baruch College (CUNY) from 1998-2001. He has been associated with Circle T Partners, LP, a \$400 million equity hedge fund; and Willowbridge Associates, a \$1 billion-plus commodities trading adviser, where his responsibilities included the ongoing refinement of a market-neutral statistically based ("stat-arb") stock selection model.  Dr. Castaldo holds a B.S. in electrical engineering/computer science and an M.S. in management from the Massachusetts Institute of Technology, and worked as a software engineer at SEI Corporation/TMI Systems, Software Research Corp. and Systems Constructs Inc. before entering the finance profession.