CHAPTER V EXPERIMENTAL RESULTS FOR CASE STUDY
It is nowadays a common notion that vast amounts of capital are traded through the Stock Markets all around the world. National economies are strongly linked and heavily influenced of the performance of their Stock Markets. Moreover, recently the Markets have become a more accessible investment tool, not only for strategic investors but for common people as well. Consequently they are not only related to macroeconomic parameters, but they influence everyday life in a more direct way. Therefore they constitute a mechanism which has important and direct social impacts. The characteristic that all Stock Markets have in common is the uncertainty, which is related with their short and long-term future state. This feature is undesirable for the investor but it is also unavoidable whenever the Stock Market is selected as the investment tool. The best that one can do is to try to reduce this uncertainty. Stock Market Prediction (or Forecasting) is one of the instruments in this process. The financial market is a complex, evolutionary, and non-linear dynamical system. The field of financial forecasting is characterized by data intensity, noise, non-stationary, unstructured nature, high degree of uncertainty, and hidden relationships. Many factors interact in finance including political events, general economic conditions, and traders expectations. Therefore, predicting finance market price movements is quite difficult. Increasingly, according to academic investigations, movements in market prices are not random. Rather, they behave in a highly nonlinear, dynamic manner. In literature a number of different methods have been applied in order to predict Stock Market returns. These methods can be grouped in four major categories: i) Technical Analysis Methods, ii) Fundamental Analysis Methods, iii) Traditional Time Series Forecasting and iv) Machine Learning Methods. Technical analysts, known as chartists, attempt to predict the market by tracing patterns that come from the study of charts which describe historic data of the market. Fundamental analysts study the intrinsic value of an stock and they invest on it if they estimate that its current value is lower that its intrinsic value. In Traditional
Time Series forecasting an attempt to create linear prediction models to trace patterns in historic data takes place. These linear models are divided in two categories: the univariate and the multivariate regression models, depending on whether they use one of more variables to approximate the Stock Market time series. There are a number of methods that have been developed under the common label Machine Learning and these methods use a set of samples and try to trace patterns in it (linear or non-linear) in order to approximate the underlying function that generated the data. Finally, artificial Neural Network has taken a great prominent role in forecasting the share market and a number of researches haven done in this area. Data collection: The historical data is obtained the from the finance section of Yahoo and the National Stock Exchange. The whole data set covers a total of 676 pairs of observations. The data set is divided into two parts. The first part consists 640 pairs of observations are used to determine the specifications of the models and parameters. The second part contains 36 pairs of observations which are reserved for evaluation and comparison of performances among forecasting models. Model 1: Back propagation Model: In this section, an empirical study on building a stock buying or selling is predicted using a feed-forward Back propagation network. It is trained and tested with past price data from National Stock Exchange. A multilayer perceptron (MLP) with a sufficient number of hidden nodes can, in principle, approximate any continuous nonlinear function. This means that if such patterns exist, a BPN can be trained to discover the patterns provided that: A sufficient amount of appropriate data is available to train the BPN. An appropriate network topology is selected. There is a sufficient number of hidden nodes. An appropriate learning rate and momentum are used.
The network consist of consists of N input nodes, K hidden nodes and M output nodes. Let opm and opk be the output of output node m and hidden node k from input pattern p, respectively. Assume and output node m, and
nk km
is the network weight for hidden node k
is the network weight for input node n and hidden node
k. Also, let xpn be the input value in input node n for input pattern p, and tpm be the target output value in the output node m for input pattern p. Note that the symbol represents the difference between the current and the new value in the next iteration. The standard algorithm for BPN is as follow: Step 1: Initialization: Initialize all weights and refer to them as current weight
km
(0) and nk. Set the learning rate and the momentum factor to
small positive values (e.g. 0.1). Set the error threshold E and the iteration number i = 0. Step 2: Forward Pass: Select the input pattern xp = { xp1, , x pn} from the training set and compute opm (i) and opk . Use the desired target tp = { tp1,y, tpm } associated with xp to compute the sum of the squared system error, E(i), for all input patterns. If E(i) <= E, then the algorithm is completed and the convergence is achieved; otherwise, go to step 3 (Backward Pass). Step 3: Backward Pass: Compute the changes of the weights for the next iteration BPN system: Defining the output goal. Determining the proper input data and necessary preprocessing. Choosing the network architecture and specifying the training algorithm. Determining the training and testing strategy. Determining the optimum network topology. Evaluating the results.
km
(i +1) and
nk
(i +1).
In general, the following issues need to be considered during the design of a
The major benefit of the BPN is that it is capable of learning the nonlinear mapping between the inputs and outputs by using an appropriate network topology and given a sufficient amount of training data. Once the underlying mapping has been learnt sufficiently well, the BPN can also produce an acceptable output for some unseen data. Obviously, this feature makes the BPN an attractive candidate for prediction tasks. The major presumption of technical analysis is that history repeats itself and that trends and patterns exist in the price data. A traditional technical analyst tries to identify these patterns by examining price charts and technical indicators. If
such patterns are identified, he or she can presumably predict the future price movements. A single output, which is the buy or sell trading signal. The basic input data includes raw data such as the daily open, high, low and close prices, and trading volumes and in addition, the following is also be considered : Fundamental factors P/E ratios and dividend yield, dividend payment; term. The training and testing strategy is probably the most important issue in designing a BPN. A moving-windows approach, as described in Skabar and Cloete (2002)[50], was adopted for training and testing. The disadvantage of this BPN is care must be taken in choosing the appropriate size of the training set. On one hand, if the training set is too small, will converge easily, but it will not predict with acceptable accuracy because the network is unable to learn the underlying patterns sufficiently well with such a limited amount of data. On the other hand, if the training set is too large, it will be more difficult for the network to converge. In fact, sometimes it fails to converge at all. Moreover, even if the network can converge, it may have learned some historical patterns that may no longer be effective because the market conditions have already changed too drastically. Model 2: Radial Basis Neural Network Technical indicators - n-day moving averages and momentum, among others; Interest rates - both long and short
[50] Skabar, A., Cloete, I., 2002. Neural networks, financial trading and the efficient market hypothesis. Australian Computer Science Communications 24 (1), 241249.