Forecasting of Stock Market Price Using Neural Networks

Mudesira Munir
Managers must realize and accept the limitations of forecasting and the uncertainty associated with all forms of predictions; otherwise, unexpected results and unpleasant surprises may await them. As long as the patterns or relationships do not change, forecasting will be accurate. The selection and implementation of a proper forecasting methodology has always been a planning and control issue for most firms and agencies. This thesis will suggest and describe the application of Artificial Neural Networks for forecasting of stock market closing prices.


Chapter #1

1.1 STOCK MARKET A stock market is a public for the trading of company stock and derivatives at an agreed price; these are securities listed on a stock exchange as well as those only traded privately. The stock market has been one of the most popular investments owing to its high returns. The stocks are displayed and traded on stock exchanges which are entities of a corporation or mutual organization specialized in the business of bringing buyers and sellers of the organizations to a listing of stocks and securities together. The major stock market in the America is NYSE while in Canada; it is the Toronto Stock Exchange (TSX). ‡ Trading Participants in the stock market range from small individual stock investors to large hedge fund traders, Actual trades are based on an auction market model where a potential buyer bids a specific price for a stock and a potential seller asks a specific price for the stock. Buying or selling at market means you will accept any price (Bid or Ask) for the stock, respectively. When the prices (Bid or Ask) match, a sale takes place, on a first come first served basis if there are multiple bidders or askers at a given price. The stock exchanges are builtin order to facilitate buyers and sellers, thus providing a marketplace. The stock exchanges provide Real Time trading system and information for the listed securities. ‡ The behavior of the stock market Positive or upward trends are referred to as bull markets; negative or downward trends are referred to as bear markets. Over-reactions may occur so that excessive optimism may drive prices unduly high or excessive pessimism may drive prices unduly low. Forecasters continue to debate whether financial markets are generally efficient (EM hypothesis). According to one understanding of the efficient market hypothesis (EMH), only changes in fundamental factors, such as the viewpoint for margins, profits or dividends, ought to affect share prices beyond the short term, where random noise in the system data

may prevail. But this viewpoint known as hard EMH also predicts that little or no trading should take place, contrary to fact, since prices are already at or near equilibrium, having priced in all public knowledge. The hard efficient-market hypothesis is sorely tested by such events as the stock market crash in 1987, when the Dow Jones index plummeted 22.6 percent the largest-ever one-day fall in the United States. 1.2 STOCK MARKET INDICES Stock market indices are classified in many ways. A global stock market index includes typically large companies without regard for where they are traded. The examples are MSCI World and S&P Global 100. Each country or nation has its own stock market index representing its performance. The most commonly quoted market indices are national indices composed of the stocks of large companies listed on a nation's largest stock exchanges, such as the American S&P 500, the Japanese Nikkei 225, and the Karachi KSE 100. The idea may be extended well beyond an exchange. The Dow Jones Total Stock Market Index represents the stocks of nearly every publicly traded company in the United States, including all U.S. stocks traded on the New York Stock Exchange (NYSE) and mostly traded on the NASDAQ and American Stock Exchange (AMEX). More specialized indices exist tracking the performance of specific sectors of the market. 1.3 INDEX VERSIONS The indexes, such as the S&P 500, have many versions. These versions can vary based on how the index components are weighted and on how dividends are accounted for. For example, there are three (3) versions of the S&P 500 index: price return, which only considers the price of the components, total return, which accounts for dividend reinvestment, and net total return, which accounts for dividend reinvestment after the deduction of a withholding tax. 1.4 MOST POPULAR AND LARGEST STOCK MARKETS: 1.4.1 American Stock Exchange (AMEX): Today AMEX is one of the largest options exchanges in the United States. 1.4.2 New York Stock Exchange (NYSE): NYSE is founded in 1792, the Big Board is the where the big companies have shares. The NYSE is where companies such as Coca cola, McDonald's, General Electric

and Wal-Mart all trades. Whereas, America's largest companies and others, they all trade on the NYSE. 1.4.3 NASDAQ The most popular over-the-counter (OTC) market is the NASDAQ¶s stock exchange. This is a virtual exchange which has no central location or floor brokers as all trading is done through a network of dealers. For many years, the largest companies traded on the NYSE, while the second tier stocks traded on all the other exchanges. Thanks to the dot-com boom of the late 1990's, some of America's largest firms trade on the NASDAQ, including, Google, Microsoft, Cisco, Intel and Dell. 1.4.4 UNITED KINDOM: The London stock exchange plays a vital role in maintaining one of the world¶s leading financial centers. The current LSE market includes not only the ordinary shares of local and foreign companies and fixed interests securities issued by private borrowers but also the primary and secondary markets in British government securities. At the end of 2000, there were 2,428 domestic listed companies with the market capitalization of 1,812 billion pounds. 1.4.5 GERMANY: There are eight technically independent stock exchanges in Frankfurt. However they actually cooperate closely. Most recently the stock exchanges of Frankfurt, Dusseldorf and Munich extended their cooperation in order to establish uniform pricing mechanism for the DAX 100securities. Recently the stock exchanges of Frankfurt, Dusseeeld offer the DAX 100securities .There were 1088 domestic companies listed on the German stock exchange at the end of 2000. 1.4.6 CHINA: China is the most populous country in the world and while its stock market is currently rudimentary, given the recent growth and the potential growth of the Chinese economy china¶s stock market will inevitably be an important world stock market. The Chinese stock market was divided into two types A-shares a, which are restricted to Chinese investors and B-shares which were restricted to non-Chinese investors.The A-share market was dominated in chines Yuan and was much larger than

the B-shareThe B-share market was started in 1991 to give private companies and entrepreneurs access to foreign capital and give foreigners with hard currency Shanghai B-shares are quoted in U.S dollars and Shenzhen B-shares are quoted in Hong Kong dollars although these two currencies are linked. 1.4.7 JAPAN: The Japanese stock market has a history of more than 120 years, beginning with the establishment of the Tokyo stock exchange (TSE) in 1878. During the 1980 the Japanese securities market changed substantially. Since March 2000, when the Hiroshima and Niigata exchanges were merged with the TSE there have been six stock exchanges in Japan. OPERATIONS: The TSE market is a continuous market where buy and sell orders directly interact with one another. The trading of shares is carried out under the zebra method which is similar to an open outcry method. Prices are first established at the beginning of the trading session based on orders placed by regular members before the start of the trading. The central book is kept by satori members who function solely as intermediaries between the regular members. They are not allowed to trade any listed stock for their own account or to accept orders from the investing public. After the opening price is established satori members match. 1.5 FORECASTING STOCK MARKET: Recently forecasting stock market return is gaining more attention, maybe because of the fact that if the direction of the market is successfully predicted the investors may be better guided and also monetary rewards will be substantial. If any system which can consistently predict the trends of the dynamic stock market be developed, would make the owner of the system wealthy. Another motivation for research in this field is that it possesses many theoretical and experimental challenges. The most important of these is the efficient market hypothesis (EMH) which proposes that profit from price movement is very difficult and unlikely. In an efficient market, stock market prices fully reflect available information about the market and its constituents and thus any opportunity of earning excess profit ceases to exist any longer.

So it is ascertain that no system is expected to outperform the market predictably and consistently. In the approach used in this paper, the ANN will be the signal generator, and its output will be used to either initiate or exit trades. Rules to enter and exit trades are based on the strength of the ANN output signal.

1.5.1 Application of Neural Networks to Market Prediction Neural networks are very effective when lots of examples must be analyzed, or when a structure in these data must be analyzed but a single algorithmic solution is impossible to formulate. When these conditions are present, neural networks are use as computational tools for examining data and developing models that help to identify interesting patterns or structures in the data. The data used to develop these models is known as training data. Once a neural network has been trained, and has learned the patterns that exist in that data, it can be applied to new data thereby achieving a variety of outcomes. The ability of neural networks to discover nonlinear relationships in input data makes them ideal for modeling nonlinear dynamic systems such as the stock market. Various neural network con¿gurations have been developed to model the stock market. Commonly, these systems are created in order to determine the validity of the EMH or to compare them with statistical methods such as regression. Often these networks use raw data and derived data from technical and fundamental analysis discussed previously. This thesis wills overview the use of neural networks in ¿nancial markets including a discussion of their inputs, outputs, and network organization. Also, any interesting applications of theoretical techniques will be examined. For example, many networks prune redundant nodes to enhance their performance. The networks are examined in three main areas: 1. Network environment and training data 2. Network organization 3. Network performance 1.5.2 Training a Neural Network

A neural network must be trained on some input data. The two major problems in implementing this training discussed in the following sections are: 1. De¿ning the set of input to be used (the learning environment) 2. Deciding on an algorithm to train the network 1.5.3 The Learning Environment One of the most important factors in constructing a neural network is deciding on what the network will learn. The goal of most of these networks is to decide when to buy or sell securities based on previous market indicators. The challenge is determining which indicators and input data will be used, and gathering enough training data to train the system appropriately. The input data may be raw data on volume, price, or daily change, but it may also include derived data such as technical indicators (moving average, trend-line indicators, etc.) or fundamental indicators (intrinsic share value, economic environment, etc.). The input data should allow the neural network to generalize market behavior while containing limited redundant data. Determining the proper input data is the ¿rst step in training the network. The second step is presenting the input data in a way that allows the network to learn properly without overtraining. Various training procedures have been developed to train these networks. 1.5.4 Network Training Training a network involves storing input patterns in a way so that the system minimizes its error and improves its performance. The training algorithm may vary depending on the network architecture, butBack Propagation algorithm is the most common training algorithm used when designing ¿nancial neural networks. This section describes some of the training techniques and their associated challenges in some implemented systems. The most common network architecture for ¿nancial neural networks is a multilayer feed forward network trained using Back Propagation. Back Propagation is the process of backpropagating errors through the system from the output layer towards the input layer during training. Back Propagation is an important algorithm because hidden units have no Training Target value that can be used, so they must be trained based on

errors from previous layers. The Output layer op back Propagation algorithm is the only layer which has a Target value for which to compare. As the errors terms are BackPropagated through the nodes, the connection weights are changed. Training Procedure occurs until the errors in the weights are small enough to be accepted. It is interesting to note that the type of activation function (such as threshold, sigmoid) used in the neural network nodes can be a factor on what data is being learned. The major area of consideration in training a neural network is deciding when to stop training. Since the ability to generalize is elementary for these networks to predict future stock prices, overtraining is a serious problem. Overtraining occurs when the system memorizes patterns and thus loses the ability to generalize. It is an important factor in these prediction systems as their primary use is to predict or generalize on input data that itpatteronized. Overtraining can occur by having too many hidden nodes or training for too many time periods (epochs). However, overtraining can be prevented by performing test and train procedures or cross-validation. The test and train procedure involves training the network on most of the patterns (usually around 90%) and then testing the network on the remaining patterns. The network¶s performance on the test set is a good indication of its ability to generalize and handle data it has not been trained on. If the performance on the test set is poor, the network con¿guration or learning parameters can be changed. The network is then retrained until its performance is satisfactory. The quantity of training data is also important. Ideally, it is desirable to have as much training data as the system can be feasibly trained with. 1.6 Research Purpose: Testing and training of neural networks will be performed over the stock market data in order to specify a neural network¶s ability to predict future trends of Stock Market Indices. It will then analyze the use of neural network as a forecasting tool. Finally, the probability of the model¶s forecast being correct will be calculated using conditional probabilities. While only briefly discussing neural network theory, this research will determine the feasibility and practicality of using neural networks as a forecasting tool for the individual investor.

The research describes the methodology required to build, train, and test a neural network using software. In this research, an S&P500, NASDAQ, DOW Jones etc. network model was validated using available data. Finally, and potentially most importantly for the investor, model accuracy probabilities were generated. This was done by combining historical market movement probabilities with the model accuracy probability. This conditional probability could prove to be a vital tool for investment decision making. The entrance of the neural network as an investment tool for the individual investor was one of the many things brought about by the explosive growth personal computers. . It was concluded that neural networks do have the capability to forecast financial markets and, if properly trained, the individual investor could benefit from using this forecasting tool.


Forecasting of stock market returns is an important issue in finance. Nowadays artificialneural networks (ANNs) have been popularly applied to finance problems such as stockexchange index prediction, bankruptcy prediction and corporate bond classification. AnArtificial Neural Network model is a computer model whose architecture essentially mimics the learningcapability of the human brain. The elements of an Artificial Neural Network resemble the biological structure of neurons and the internal operation of a human brain. Many simpleinterconnected elements that may use threshold or sigmoid function are operating in parallelprocessing at multiple layers. In some applications it has been specified that ANNs havelimitations for learning the data patterns. They may perform inconsistently andunpredictable because of the complex financial data used. Sometimes data is sovoluminous that learning patterns may not work. Continuous and large volume of dataneeds to be checked for redundancy and the data size should be decreased for the algorithm to work in a shorter time and give more generalized solutions There are several motivations for trying to forecast stock market prices. The most basic of these is ¿nancial return. Any software system that can constantly pick winners and losers in the dynamic market place would make the owner of the system very wealthy. Thus, many individuals including researchers, investment professionals, and average investors are continually looking for a software system which will yield them high returns. There is another motivation in the research and ¿nancial communities. The Efficient Market Hypothesis effectively states that no system can continually beat the market because if this system becomes public, everyone will use it, thus negating its potential gain. There has been an ongoing debate about the validity of theEfficient Market Hypothesis, and some researchers attempted to use neural networks to validate their claims. There has been no consensus on the Efficient Market Hypothesis validity, but many market observers tend to believe in its weaker forms, and thus are often unwilling to share proprietary investment systems. [16]

Neural networks (ANN¶s) are used to predict stock market prices because they are able to learn nonlinear mappings between inputs and outputs. Contrary to the Efficient Market Hypothesis, several researchers claim the stock market and other complex systems exhibit chaos. Chaos is a nonlinear deterministic process which only appears random because it cannot be easily expressed. With the artificial neural networks (ANN) ability to learn nonlinear, chaotic systems, it may be possible to outperform traditional analysis and other computer-based methods. [16] Neural networks can be used to learn to predictfuture events based on the patterns that have been observed in the historical training data;learn to classifyunseen data into pre-defined groups based on characteristics observed in the training data;learn to clusterthe training data into natural groups based on the similarity of characteristics in the training data.

Back propagation algorithm The Back propagation algorithm seeks to minimize the error term between the outputs of the neural net and the actual desired output value. The error term is calculated by comparing the net output to the desired output and is then fed back through the network causing the synaptic weights to be changed in an effort to minimize error. The process is repeated until the error reaches a minimum value. [6] In a forward pass the outputs are calculated and the error term at the output units calculated. And in a backward pass, the output unit error is used to alter weights on the output units. Then the error term at the hidden nodes is calculated by backpropagating the error at the output units through the weights, and the weights on the hidden nodes altered using these values.

2.2 BACK PROPAGATION LEARNING ALGORITHM The major steps of the back propagation learning algorithm are given below: 1. Input data to be train. 2. Hidden nodes calculate their outputs. 3. Output nodes calculate their outputs on the basis of Step 2. 4. Calculate the differences between the results of Step 3 and targets. 5. Apply the first part of the training rule using the results of Step 4. 6. For each hidden node in a hidden layer, n, calculate d(n). (derivative) 7. Apply the second part of the training rule using the results of Step 6. Steps 1 through 3 are often called the forward pass, and steps 4 through 7 are often called the backward pass. For each pair ofdata to be learned a forward pass and backwards pass is performed. This is repeated over and over again until the prediction error is at a low enough level.

Figure 2.1: Flow Graph for the Neural Network

2.3 Calculations and Transfer Function:
The behavior of a NN (Neural Network) depends on both the weights and the input/output function or transfer function that is specified for the units. This function typically falls into one of three categories:
y Linear

For linear units, the output activity is proportional to the total weighted Wij output.
y Threshold

For threshold units, the output are set at one of two levels, depending on whether the total input is greater than or less than some threshold value.
y Sigmoid

For sigmoid units, the output varies linearly but not continually as the input changes. Sigmoid units accept a greater resemblance to real neurons than do linear or threshold units, but all three must be considered rough approximations.

The network uses Equation 1 to update the weight Wij from a given node i N to the current node j N ; where t refers to the number of times the network has been updated and µl¶ refers to the learning parameter. The learning parameter, or learning rate, controls the rate the weight is changed as learning takes place. The sensitivity of node j N to a change in weight Wij is represented by W ije and will be defined more fully below. Ii = ™WijOi + Oi= 1 / 1 + e-i OUTPUT LAYER: Errj = Oj(1-Oj)(Ij ± Oj) HIDDEN LAYER: Errj = Oj(1-Oj)™ErrWij Also , Bias and weight changes are as follows

¨ = L Errj ¨Wij = L ErrjOi

Figure 2 .2:Back Propagation

CHAN Man-Chung, WONG Chi-Cheong, LAM Chi-Chung [8] collect daily trading data of eleven companies in 1994-1996 from Shanghai Stock Exchange for technical analysis of stock price. The first 500 entries were used as training data. The remaining 150 were testing data. The raw data is preprocessed into various technical indicators to gain insight into the direction that the stock market may be going. Ten technical indicators were selected as inputs of the neural network. [8] The training phase of a backpropagation network is anunconstrained nonlinear optimization problem. The goal ofthe training is to search an optimal set of connectionweights in the manner that the errors of the network outputcan be minimized. [8] Besides popular steepest descent algorithm, conjugategradient algorithm is another search method that can beused to minimize network output error in conjugatedirections. [8] Backpropagation is a difficult technique. It runs therisk of being trapped in local optimum. The starting node of the connection weights becomes an important issue toreduce the possibility of being trapped in local optimum.Random weight initialization

does not guarantee to generatea good starting point. It can be enhanced by multiple linearregressions. In this method, weights between input layerand hidden layer are still initialized randomly but weightsbetween hidden layer and output layer are obtained bymultiple linear regression. [8] EMA is a trend-following tool that gives an average value of data with greater weight to the latest data. Difference of EMA can be considered as momentum. RSI is an oscillator which measures the strength of up versus down over a certain time interval. High value of RSI indicates a strong market and low value indicates weak markets. MACD, a trend-following momentum indicator, is the difference between two moving average of price. In order to avoid difficulty in getting network outputs very close to the two endpoints, the indicators were normalized to the range of [0,1], before being input to the network. [8]. Empirical results have shownthat Neural Networks outperform linear regression. Since stock markets are complex, nonlinear, dynamic andchaotic. Neural networks are reliable for modelingnonlinear, dynamic market signals. [8] BirgulEgeli, MeltemOzturan, and BertanBadur [9] performed research on Turkish stock market. Network architecture is chosen is Multi-Layer Perceptron and Generalized Feed Forward networks. Testing and training is performed using these two network architectures. The results are then compared with the results of moving averages (MA) for 5 and 10-day time periods, which showed that artificial neural networks (ANN) have better performances than moving averages. [9]

The ANNs can effectively be used to predict the Stock Exchange using index values S&P 500 Index. Supervised learning models have been utilized in which certain output nodes were trained to respond to certain input patterns and the changes in connection weights due to learning caused those same nodes to respond to more general classes of patterns.

2.5.1 DATA SETS Experimental data were gathered by using commercial software for USA starting from March 10, 2010 to April 20, 2010. From this data set, the first 7 cases (about 90%) were taken as training and remaining as testing examples. 2.5.1NETWORK PARAMETERS For the system model described before, Artificial Neural Network model is applied with random number of hidden layers (For e.g. HL = 1, 2, 4) for Learning rate input by the investor /user for the data set. Momentum The momentum parameter controls how much of the previous weight adjustment isapplied to the current weight adjustment. As larger values are used for momentum, thegreater the influence of the current correction term, relative to previous correction terms.Momentum can also be used to provide a smoothing effect for weight adjustments [15]. Training Rate Training rate also known as Learning Rate determines the amount of the correction termapplied to adjust neuron weights. A small value for training rate tends to increaselearning time and decrease the probability of overshooting an optimal solution. A smallvalue also increases the likelihood of being stuck in local minima. Large values fortraining rate increase the chance of no learning occurring at all. [15]

In this study, ANN models were applied to the system model, using an ANN software package. ANN models¶ performances can be measured by the coefficient of determination (R2) or the mean relative percentage error. This coefficient of determination is a measure of the accuracy of prediction of the trained network models. Higher R2 values indicate better prediction. The mean relative percentage error may also be used to measure the accuracy of prediction through representing the degree of scatter. [9]

The selection of technical variables for input to the technical neural network is influencedby the technical variables discovered during the formal literature reviewStock market indicators such as moving averages, stochastic oscillators, RSI etc are used By Nikhil BakshiStock [12]. The ANN performances can be compared with Moving Averages (MA) approach for example. The moving average is the average of lagged index values over a specified past period ------------- (25 and 50 days in this study). The mean relative percentage errors were calculated as -----------days.

Using historical data, a neural network can be learned that describes howtrading signals are related to subsequent price movements. Thetrained neural network can then be applied to new data to predict future pricemovements and make investments. [12] 2.8.1 Hidden Layer One hidden layer is used with a configurable number of neurons. A regularsigmoid function is used as a transfer function. 2.8.2 Output The output is the expected price change in a window of days. The value is inthe continuous [0,1] range with 1.0 representing a 10% rise, 0.5 representingno price change and 0.0 representing a 10% price drop.

2.8.3 Training The neural network's weights are learned using the backpropagation algorithm with a configurable learning rate and number of epochs. [12] The input variables were the highest and lowest prices paidduring the day, the closing price. The output variable used in thisstudy was prepared according to the direction of the index onthe following day. The closing price is first normalized by the investor using Min-Max Normalization using the scale [0, 1]. Then data is loaded for the input of Neural Network.

For the purposes of this study, all data relating to an index(the highest and lowest prices paid during the day and theclosing price) was obtained from using the commercially available software the NEURO Stock master. [10]

CHAPTER # 3 RESEARCH METHODOS / DESIGN 3.1 The Research Questions
The following research questions allow the research to meet the objectives proposed: y y y What is the mathematical theory behind the Back propagation neural network? Can neural networks accurately forecast a stock market index? Can neural networks be used as a practical forecasting tool by individual

investors? Conduct a literature review of books magazines and World Wide Web on the topic of neural networks Identify the mathematical theory of the back propagation neural networks. . Select a Stock Index to make forecasts upon. Determine what to forecast and the future point for the forecast variable Days. Determine the inputs to the neural network assemble the historical input data and preprocesses it using MS Excel.


Figure 3.1 Data Flow Diagram


Figure 3.2 Database Design


Neuron +ThresholdValue +LastInputValue +SigmoidFunction() +SigmoidfunctionPrime() +GetSumOfInputWeigths() +Output() * * Connection -Weight -PreviousWeight *

1 1 Layer -LastOutputValues +Output * NeuralNetwork +MaxIterations +LearningRate +Momentum +RandomMin +RandomMax +Learn() +Output() +Randomize() -CalculateErrors() -UpdateWeights()

Figure 3.3 Class diagram of the system

Neuron class represents the artificial neuron, the basic structure of a neural network. It carries a threshold value, has input values, uses a sigmoid function and combination of those properties are used to calculate the overall output. Neurons are connected to each other with connection lines which carry weights, represented by the Connection association class. Neurons reside in layers, so the Layer class represents the

artificial layer where a layer contains one to many neurons and a neuron can only reside in one layer. Therefore, the relationship between a neuron and a layer is one-to-many. The neural network, which is represented by the Neural Network class, has parameters like Max Iterations, Learning Rate, and Momentum and consists of several layers. Therefore, the relationship between Neural Network and Layer classes is one-to-many. The training algorithm is backpropagation and the threshold function used in artificial neurons is the sigmoid function. The program is developed as a library package so that it can easily be used.


Figure 3.4 Use Case Modeling of software designed

In addition to stock market prediction, neural networks have been trained to perform a variety of ¿nancial related tasks. There are experimental and commercial systems used for tracking commodity markets and futures, foreign exchange trading,

¿nancial planning, company stability, and bankruptcy prediction. As the application of neural networks in the ¿nancial area is so vast, this paper will focus on stock market prediction. In stock trading, there are typically two methods used to determine whether to buy or sell a stockfundamental analysis and technical analysis. Fundamental analysis in forecasting stock market prices is used to evaluate the intrinsic value of the underlying asset of the stock by examining related economic and company-specific information. Technical analysis is used to forecast and identify patterns in stock prices from historical prices and the traded volume of the stock. Due to the widely available dataset of accurate historical prices (open, high, low and close) and the volume traded, this project will focus on these as inputs to the Neural Network. The following indicators, which are derived from historical prices and commonly used in technical trading, are also included as inputs to the NN: simple moving average, exponential moving average, moving average convergence divergence, relative strength index, and stochastic oscillator. Finally, although neural networks are used primarily as an application tool in the ¿nancial environment, several research improvements have been made during their implementation. Notable improvements in network design and training and the application of theoretical techniques are demonstrated by the examination of several example systems. 3.6. ANALYTICAL METHODS Before the advent of computer, traded stocks and commodities primarily on intuition. As the level of investing and trading grew, people searched for tools and methods that would increase their gains while minimizing their risk. Statistics, technical analysis, fundamental analysis, linear regression is all used to attempt to predict and bene¿t from the market¶s direction. None of these techniques has proven to be the consistently correct prediction tool that is desired, and many analysts argue about the usefulness of many of the approaches. However, these methods are presented as they are commonly used in practice and represent a base-level standard for which neural networks should outperform. Also, many of these techniques are used to preprocess input data, and their results are fed into neural networks as input. [11]

3.6.1 FUNDAMENTAL ANALYSIS Fundamental analysis involves the in-depth analysis of a stock company¶s performance and portability to determine its share price. A method of evaluating a security level that entails attempting to measure its intrinsic value by examining related economic, financial and other qualitative and quantitative factors. Fundamental analysts attempt to study that can affect the security's value, including macroeconomic factors (like the overall economy and industry conditions) and company-specific factors (like financial condition and management).

The goal of performing fundamental analysis is to produce a value that an investor can compare with the security's current price, with the aim of figuring out what sort of position to take with that security value (underpriced = buy, overpriced = sell or shortnotpriced This Fundamental analysis is considered to be the opposite of technical analysis. The advantages of fundamental analysis are its systematic approach and its ability to predict changes before they show up on the charts. Companies are compared with one another, and their growth prospects are related to the current economic environment. This allows the investor to become more familiar with the company. Also, it is hard to time the market using fundamental analysis. Although the outstanding information may warrant stock movement, the actual movement may be delayed due to unknown factors or until the rest of the market interprets the information in the same way. However, fundamental analysis is a superior method for long-term stability and growth. Basically, fundamental analysis assumes investors are 90% logical, examining their investments in detail, whereas technical analysis assumes investors are 90% psychological, reacting to changes in the market environment in predictable ways. [11] =hold).

3.6.2 TECHNICAL ANALYSIS The idea behind technical analysis is that share prices move in trends dictated by the constantly changing attitudes of investors in response to different forces. Using price,

volume, and open interest statistics, the technical analyst uses charts to predict future stock movements. Technical analysis rests on the assumption that history repeats itself and that future market direction can be determined by examining past prices. Thus, technical analysis is controversial and contradicts the Ef¿cient Market Hypothesis. However, it is used by approximately 90% of the major stock traders. Despite its widespread use, technical analysis is criticized because it is highly subjective. Different individuals can interpret charts in different manners. Price charts are used to analyze trends. Trends are assumed to be based on supply and demand issues which often have cyclical or wave patterns. There are a variety of technical indicators derived from chart analysis which can be formalized into trading rules or used as inputs to neural networks. Some technical indicator categories include ¿lter indicators, momentum indicators, trend line analysis, cycle theory, volume indicators, wave analysis, and pattern analysis. Indicators may provide short or long term information, help identify trends or cycles in the market, or indicate the strength of the stock price using support and resistance levels. An example of a technical indicator is the moving average. The moving average averages stock prices over a given length of time allowing trends to be more visible. Several trading rules have been developed which pertain to the moving average. For example, "when a closing price moves above a moving average a buy signal is generated."[11]. unfortunately, these indicators often give false signals and lag the market. That is, since a moving average is a past estimate, a technical trader often misses a lot of the potential in the stock movement before the appropriate trading signal is generated. Thus, although technical analysis may yield insights into the market, its highly subjective nature and inherent time delay does not make it ideal for the fast, dynamic trading markets of today.

3.7 Classification of Indicators
Indicators are mathematical/statistical functions that are applied over stock properties such as close, high, low and volume. These Indicators are broadly classified into the following important categories:

y y y y y

Market Momentum Indicators Market Volatility Indicators Market Trend Indicators Broad Market Indicators General Momentum Indicators

The following are some of the basic indicators that are used for creating the baseline model. These indicators are known to provide useful information for forecasting using neural networks. 3.7.1 Relative Strength Index Relative Strength Index is a measure of the strength that is intrinsic in a field and is calculated using the amount of upward and downward price changes over a given period of a time. It has arrange of 0to 100 with values typically remaining b/w 30 and 70 .The formula for computing Relative strength Index is as follows RSI=100-[100/ (1+RS)] Where RSI=Relative strength Index RS=Average of x days up closes average of x days closes. In addition, the value is defined 100 when no downward price changes occur during the period of calculation. y Relative Strength Index usually leads by the price by forming peaks and valleys before the price data, especially around the values 30 and 70. y When the RSI diverges from the price, the price eventually follows a corrective trend towards the direction of the Index. 3.7.2 Money Flow Index Money Flow Index is a measure of the strength of the monetary investment flowing into or out of a stock traded in the open market. It is principally derived by comparing the volume of upward and downward price changes over a given period of time. The money flow Index is based on the quantity of Money Ratio, which is the ratio of positive money flow to negative money flow over the given period of time. Money Flow=Typical Price * Volume Money Ratio= Positive Money Flow Negative Money Flow MFI=100 100

1+Money Ratio Positive Money Flow is defined as sum of prices multiplied by the volume on days when the price increases. Negative Money flow is defined similarly, except that it includes only days when the prices decreases. The Money Flow Index typically has the range of 0 to 100 with values rarely exceeding the bounds 20 and 80. 3.7.3 Stochastic Oscillator The stochastic Oscillator may be defined as a measure of the difference between the current closing price of a security and its lowest low price, relative to its highest high price for a given period of a time. The value is a percent rating for the closing price, relative to the trading range b/w its recent highest and lowest prices. A value of zero indicates that the security had a closing price at its lowest recent low. A value of 100 indicates it that the security had a closing price at its highest recent high. The value is often smoothed using a slowing period to eliminate noise in the trend graph. It is used in technical analysis. Interpretation The idea behind to use this indicator is that prices tend to close near their past highs in bull markets, and near their lows in bear markets. Transaction signals can be spotted when the stochastic oscillator crosses its moving average. Two stochastic oscillator indicators are typically calculated to assess future variations in prices, a fast (%K) and slow (%D). Comparisons of these statistics are a good indicator of speed at which prices are changing or the Impulse of Price. %K is the same as Williams %R, though on a scale 0 to 100 instead of -100 to 0, but the terminology for the two are kept separate. The fast stochastic oscillator or Stoch %K calculates the ratio of two closing price statistics: the difference between the latest closing price and the lowest price in the last N days over the difference between the highest and lowest prices in the last N days:


CP is closing price LOW is low price HIGH is high price

3.7.4 Moving average Convergence /divergence (MACD) The MACD is the difference b/w the short and long term moving averages for a field .The MACD is generally a specific instance of a value Oscillator and is mostly used on the closing price of a security to detect price trends .When the MACD is on an increasing trend, prices are trending higher. If the MACD is on decreasing trend, prices are trending lower.

3.7.5 Moving Averages: Moving Average is calculated by averaging together the previous values over the given period, including the current value. MA = n Moving averages are useful for eliminating noise in raw data. Analysis of the moving average of the price yields a more general picture of the underlying trends. 3.7.6Simple moving average A simple moving average (SMA) is the unweighted mean of the previous n data points. For example, a 10-day simple moving average of closing price is the mean of the previous 10 days' closing prices. If those prices are formula is then the Closing Price

When calculating successive values, a new value comes into the sum and an old value drops out, meaning a full summation each time is unnecessary,

Perhaps the simplest of all Time series forecasting techniques is a moving average. To use this method, we calculate the average of, say, three periods of actual demand and use that to forecast the next period's demand. If this three-period average is to be used as a forecast, it would have to forecast demand in a future period. [14] Because each average moves ahead one period each time, dropping the oldest value and adding the most recent, this procedure is called a moving average. The number of periods to use in computing the average may be anything from 2 to 12 or more, with 3 or 4 periods being common. If the time series is such that there is no upward or downward trend, then the moving average is a satisfactory technique. If, how-ever, there is any trend or any seasonal effect, then the moving average will not work very well. Moving averages lag behind any trends.

Forecasting techniques (APPLIED IN THIS THESIS) can be divided into two categories: qualitative and quantitative. The former, which involve numbers, uses methodology that is not mathematical. Qualitative techniques rely on judgment, intuition, and subjective evaluation. Among the major techniques within this category are market research (surveys), Delphi (panel consensus), historical analogy, and management estimation (guess).

3.8.1Qualitative Techniques
Commercial packages are also a reference for comparison of the results of the software developed. Several commercial neural network packages are widely used in Artificial Neural Network studies, one of which is Neuro Stock master1.3, for USA. This section discusses the techniques used to develop the neural network and multiple regression financial forecasting models. The combination of these models, with

historical and conditional probabilities, will allow the investor to make decisions based on probabilities of success. The presentation first discusses the development of the models, followed by the historical probability calculations and concludes with the conditional probability calculation. The information to be forecast was first identified. Three separate indexes were chosen, the first was the S&P 500 and predicted on last 15 bars. Accurate forecast is possible when predicting the percent change versus the actual closing value of a stock or index. For example, if the S&P500 index is calculated from date 22-07-08 to 04-05 10(chart #1) Open value in 25-01-10 was 1092.4 and closing value 1096.78. In moving simulation, prediction is done while moving the target learning and prediction periods. For example, initially the system is trained on data from January, tested on data for February, and then used to predict data for March. In the next iteration, the system trains on the February data, tests on the actual March data, and then predicts data for April. In this way, the system is continually updated based on the most recent data and its performance is improved.

The first neural network model, named S&P 500 Index *SPX the inputs to the model where developed by first downloading the following raw data from Dial Data using Data Downloader software: y y y S&P 500 Index NASDAQ Dow Jones Industrial Index

The data set encompassed the different trading days. The data was charted using stock charting software called STOCK NEUROMASTER 1.33. This established date integrity within each index. Using Microsoft Excel, each index text file was combined into one spreadsheet. The resulting file was formatted such that each index and its date were in a separate column with each row representing a trading day.

An example of a partial data set can be found in Appendix B. Date integrity is extremely important to the neural network. To ensure integrity was maintained, each index¶s date column was checked for integrity against the S&P500 date using a simple ³if then´ rule in Excel. If the dates were equal, it placed a zero in a row. If the dates were not equal, Excel placed a 1 in a row. This function was copied down the entire data set and then summed at the bottom. discrepancies. Prediction statistics for buy, sell and hold are also tabulated separately [APPENDIX A] y y y S&P 500 Index NASDAQ Dow Jones Industrial Index A sum of zero represented no date integrity

This (CHART) represents the future closing price of the S&P500 Index and was the actual value the network used to compare against its prediction during training. This future information was created by simply copying the S&P 500 closing price into the last column of the spreadsheet deleting the first 10 days of data and moving the remaining data up 10 rows. Once this was complete, all date columns were deleted with the exception of the one associated with the S&P 500 data. The file was saved as an Excel 4 worksheet so it could be imported into NEUROSTOCK MASTER. .

The following data was imported into the neural network software package as a pattern file: S&P500 O S&P500 H NASDAQ O NASDAQ H DOW Jones O DOW Jones H

S&P500 L S&P500 V S&P500 C


DOW Jones L DOW Jones V DOW Jones C

3.8.2 Quantitative Technique:
When attempting to forecast demand for a new item, one faces a shortage of historical data. If the there is no special event data is very similar, quantitative techniques may be used. But if the relationship is vague, it may be more appropriate to relate the products only qualitatively in order to get an impression of demand patterns or aggregate demand. [1]

Software Tools required for the implementation of the software: 1. Microsoft Visual C # 2005 Express Edition 2. Framework 2.1.2 version 3. Microsoft Access 2003-2007 4. Microsoft Windows XP 2007

Include Libraries:
y y y AFoge.Neuro.dll AForge.Controls.dll AForge.dll

Include Sources:
y y y Core Control Neuro

y y BackPropagation Time series

Figure 4.1 Inputvariables to ANN

Figure 4.2 Stock Quotes Database View

Figure 2.3 Normalization

Figure 2.4 Money Flow Index

Figure 2.5 Stochastic Oscillator

Figure 2.6 Relative strength indexes

The data collected by using an available software as a sample for S&P 500, Nasdaq¶s and Dow Jones for open, high, low, close and volume .Buy sell is also indicated in separate tables .The bar graphs are also plotted in order to analyze time series forecasting using the software designed in order to validate the ability of Neural Networks ability to forecast the stock Market Prices.

S &P 500:

Table 4.1


Table 4.2

Table 4.3


Table 4.4

Table 4.5

Graph 4.1

Graph 4.2

Graph 4.3

Chapter # 5

The learning rate (training rate) was set to ---0.1&0.2---and the momentum was set to ---0 or 0.1----. Within the Training Criteria module, pattern selection was set to random, calibration interval was set to every random patterns and stopping criteria was set to stop training at --1000--events since minimum average error within the test set. Within the training module, the network was set to automatically save the weights for the best test set. The network was then trained within the Training module. The trained network was then applied to the ----2----sample test patterns. A testing period to find a suitable model for the problem area is observable due to the lack of methodologies for use in the development of an Artificial Neuron Network Model. An intuitive Approach is needed for the architecture in order to develop a suitable model and develop parameters relating to the training statute. Backpropagation seems fit for the time series Analysis and is an effective model in the prediction of Stock Exchange Market S&P 100 Index. Training was not realized in some of the ineffective models, with a resulting failure to decrease the percentage of the error in the network. In some cases, the network memorized the data, resulting in an observable tendency to decrease the percentage of error in the training data, while increasing the percentage of error in the test data. All ineffective models were eliminated, and as a result, the model, parameters given above, found the training data to be----80 to 88----% in 2 Or 3 iterations.

No of testing Data Points No of correct change predictions No of incorrect change predictions Percentage Accuracy Learning Rate & Momentum Table 5.1

25 22 3 88% 0.1 & 0

No of testing Data Points No of correct change predictions No of incorrect change predictions Percentage Accuracy Learning Rate & Momentum Learning Error Table 5.2

45 38 7 84 % 0.2 & 0.1 0.268

Everybody wants to know whether the stock market will be up or down tomorrow. Marketing analyst want to predict future sales. Economists want to predict economic cycles. Banks can use neural networks to scan credit and loan applications to estimate bankruptcy probabilities, while money managers can use neural networks to plan and construct pro¿table portfolios in real-time. It is to expect that µTime series Forecaster´ can be improved to make predictions with greater accuracy in the following ways:  Including more selective fundamental factors, technical indicators, and broader market indices into the input data set.  Fine tuning and proper training the network topology and learning parameters. This thesis has attempted to analyze the Artificial Neural Networks ability to forecasts stock market prices. In doing so, it has laid down a structured approach to creating trading systems using neural networks, and has defined the key characteristics necessary to benchmark these trading systems tested on out-of-sample data. Backpropagation neural network is employed during this thesis, focusing on different sets of input variables, and they were trained and tested over different portions of the market. The purpose of this section of the thesis is to provide a discussion on why the neural networks achieved the results they did, and then to formally draw conclusions about the research question and thesis hypotheses from the results. Another possible explanation is that the neural network did not perform poorly, and that it gave good signals, however, these signals were not focused enough to allow a trading system to profit. This implies that when the market is rising, many stocks tend to benefit and participate in this rise. By implication, when the overall market is listless for a prolonged period, any strategy which relies on exploiting abnormal returns has limited its scope for success. Function profiles are excellent tools to visually inspect the profile of any mathematical function. They allow a deep and visual understanding of functions characteristics, and provide the ability to visually assess the function profile in varying timeframes

simultaneously. Their main limitation becomes apparent when considering the function profile for the poorly performing Artificial Neural Networks. This limitation is that the function profile is not contextual. In other words, for a given output strength forecast by the Artificial Neural Networks, the average return displayed in the function profile is the average return of all theobservations for that output strength. This thesis has effectively attempted to document and explain the forecasting stockmarket return anomalies by using separate Artificial Neural Networks trained with fundamental and technical data. It was noted in the literature review that there is a growing trend towards using the ensemble approach to analysis amongst soft-computing researchers.


[1]. Spyros Makridakis& Steven C. Wheelwright, Forecasting Methods for Management, Fifth Edition,. John Wiley & Sons [2]. Satish Kumar, NEURAL NETWORKS A classroom approach, International Edition by McGraw Hill. [3].Application of Back-Propagation neural network in data forecasting.ppt searched from http// [4]. Thesis by Jason E.Kutsuralis, Forecasting financial Market using Neural Network, September 1998. [5]. An Empirical Methodology for Developing Stockmarket Trading Systems using Arti_cial Neural Networks By Bruce J. Vanstone_ Gavin Finniey [6]. Haykin, Simon, Neural Networks: A Comprehensive Foundation, Macmillan College Publishing Company, New York, New York, 1994. [7]. [8]. CHAN Man-Chung, WONG Chi-Cheong, LAM Chi-Chung, Financial Time

Series Forecasting by Neural Network Using Conjugate Gradient Learning Algorithm and Multiple Linear Regression Weight Initialization [9]. BirgulEgeli, MeltemOzturan, BertanBadur Stock Market Prediction Using Artificial Neural Networks [10].By BirolYildiz, Abdullah Yalama, and MetinCoskun Forecasting the Istanbul Stock Exchange National 100 Index Using an Artificial Neural Network [11].Robert J. Van Eyden. The Application of Neural Networks in the Forecasting of Share Prices.Finance and Technology Publishing, 1996. [12].By Nikhil Bakshi Stock Market Prediction Using Online Data: Fundamental and Technical Approaches ETH Zurich, August 2008 [13]. [14]. [15] By Bruce James Vanstone Trading in the Australian Stock market using Artificial Neural Networks [16]Using Neural Networks to Forecast Stock Market Prices , By Ramon Lawrence

[17].Design and implementation of NN5 for Hong Kong stock price forecasting By Philip M. Tsanga,_, Paul Kwoka, S.O. Choya, Reggie Kwanb, S.C. Nga, Jacky Maka, Jonathan Tsangc, Kai Koongd, Tak-Lam Wonge. [18]. [19] ByAkinwaleAdio T, Arogundade O.T. and Adekoya Adebayo F. ³Translated Nigeria Stock Market Prices Using ANN for effective Prediction ³

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.