A New Simulation Based Market Timing Test and an Application to Technical Analysis

Abstract

The performance analysis of investment strategies results sometime in a ambiguous conclusion; It yields higher returns than a benchmark; however, the various tests do not confirm this statistically. Here, we argue that this is due to the low power of some testing procedures, such as Student t-tests for equal means. First we show that these tests require a strategy mean return twice higher than the benchmark in standard stock market conditions in order to generate p-values lower than 5%. Thus, we propose a new performance measure based on the simulation of trading positions according to Markov chains. This allow us to keep the structure of the original trading positions. Then, we re-examine the moving average based strategies of Isakov and Marti (2011), which yield an annual mean returns that lie between 10.7% and 14.6% compared with only 6.1% for the benchmark. However, standard Student t-tests fail to reject the null hypothesis of equal means; the highest t-statistic being 1.2. With the proposed test, we found that the strategies high returns can not be achieved by taking random positions with a similar structure in the market. The economical performance is thus confirmed statistically. Finally, we perform a comprehensive Monte-Carlo analysis to illustrate the proposed test superior power compared with commonly used testing procedures in the technical analysis literature.

Keywords: simulation method, testing procedure, Markov Chains, technical analysis, moving averages

JEL Classification: C12, C15, G10, G11