A LOGIT ANALYSIS OF EFFICACY OF CHARTING PATTERNS (Zastosowanie modelu logitowego w weryfikacji skutecznosci analizy formacji cenowych)
This paper presents the results of investigation of efficacy of charting patterns, which are used by chartists to forecast changes of securities' prices. Fully objective method of identification of 6 chosen patterns is described. It is similar to the approach of Lo, Mamaysky and Wang (2000) but includes some enhancements: (1) the method of stock prices noise elimination is not statistical but instead is build upon the tenets of Dow Theory which forms the basis of trend analysis in charting, (2) it includes the stage of pre-pattern trend direction check, and (3) it allows one to analyze trends formed during periods of different length. This method combined with logit model is used to verify the quality of transaction signals which arise when patterns occur. The emphasis was put on two issues: (1) the quality of the forecast of trend change, which is generated by the occurrence of each of analyzed patterns, (2) the quality of the forecast of price change following the occurrence of the pattern. The results provide some evidence that charting patterns generate poor transaction signals. Moreover the factors which according to chartists should increase the quality of forecasts associated with patterns, do not affect it or even decrease it. For example the analysis shows that the bigger the pattern is, and the longer the period of its formation is the less probable is fulfillment of the forecast associated with the occurrence of this pattern. This finding is clearly inconsistent with the content of books and papers about charting. The only exception are results concerning the efficacy of head-and-shoulders patterns. In case of these patterns it turned out that the forecast of trend reversal which is generated by their occurrence, is worth attention. Unfortunately the forecasted strength of price trend following the occurrence of these patterns tends to be overestimated which lessens the significance of good trend reversal forecast.
CEJSH db identifier