Trend strips: a new tool to analyse financial markets.

Author:Filho, Antonio Carlos da Silva

    The analysis of time series plays a key role in the modern world of science (Hamilton, 1994; Franses, 2000). This is particularly true in those fields where almost all the information we can get from the system is provided by experimentation or observation. One of these fields is finance and through the last decades a lot of techniques were developed to refine and enlarge the analysis.

    Here we are concerned with numerical time series, as that found in the stock markets, like an asset price or an index. These series can be used, among others, for two purposes: (a) to describe the system and (b) to forecast future values of the series. This paper deals with a new kind of tool, first proposed here, that could be used for these two purposes, at least. In section 2 we just reinforce the originality of this approach; in section 3 we describe the method; in section 4 we apply the method to a particular example: the Sao Paulo Stock Market Index, the Ibovespa and in section 5 we conclude with some indications of lines of research to follow from here.


    The method is new. Therefore, there is no literature yet dedicated to its study.


    Let X = {[X.sub.1], [X.sub.2], ..., [X.sub.n]} be a time series with a discrete index t (usually the time) that spans an interval from 1 to n. From this series X we can generate another time series (that we will call Y), with n-1 elements and where each element is either 0 or 1, which are, in this model, just nominal variables. The meaning of "1" is that the value of X at some position has increased (or remained the same) over the value of X at a previous position, while "0" means that the value has decreased:


    The Y series has, obviously, n - 1 elements because we construct it starting from the second position in the X series, where the value of X is [X.sub.2]. With this definition, the Y series is a sequence of "0" e "1", as can be viewed in Table 1, which is just a small piece of the entire X and Y series.

    The present paper looks for the existence of some patterns in the Y series and, if so, to find a correlation between these patterns and what is going to happens to the [X.sub.t+1] value. More specifically, we look for the probability that the [X.sub.t+1] value will increase or remains stable and the probability that Xt+1 value will decrease, according to the patterns studied in the Y series.

    The technique consists in...

To continue reading