In this research reportthe forecasting models and technical indicators are implemented and evaluatedin order to find the best combination suitable for novice and short term stocktraders.This research reportprovides an in depth analysis of most popular recent day trading strategiesincluding technical indicators which are used as an input to machine learningclassifiers in attempt to classify the selected script data into recommendedstrategy and the other methodology used is forecasting thru business intelligencetools. For this work I have used R Studio and perform a time series forecastingusing ARIMA (Auto Regressive Integrated Moving Average) and Holt Winter model.Since both these models take the close price of the trading stock and run thealgorithms on that index in order to predict future prices.In this paper data hasbeen collected from PSX website.
The historical data for the period of fiveyears since 2012 to 2017 were taken for analysis.The work performed forforecasting strategies comprises of forecasting using ARIMA and Holt WinterModel. The first strategy adapted to forecasting is ARIMA Model in which theclose price of the day is provided as an input to the model. In order tocalculate the error accuracy of the forecasted results MAPE (Mean AbsolutePercentage Error) is used the results of MAPE shows that the forecastingaccuracy of ARIMA model in case of Attock Cement Script is 6.42% and thenumerical calculations predicted an error accuracy of 2%.Similarly theforecasted values of Lucky Cement are evaluated using MAPE and the erroraccuracy for ARIMA Model is calculated to be 8.12% while the numericalcalculations represents an error of 1% due to a small size of forecastedvalues.
The forecasted Value error calculation on D.G. Stock script using MAPEshows an error of 12% while the numerical calculations represents an erroraccuracy of 2%.Besides the ARIMAforecasting another popular method of time series forecasting is taken intoaccount; Holt Winter (HW) forecasting the error calculations are done using thesame methodology like ARIMA Model.The MAPE of HWforecasting for Attock Cement is calculated to be 7.2% while the numericalcalculations shows an error accuracy of 5%. Similarly the HW results of LuckyCement script are calculated to be 6.5% while its numerical forecasted valuesshows an error accuracy of -3%.
Holt Winter MAPE results of D.G. Cement Scriptshows an error of 14.6% while the numerical error calculations represent anerror of -1%.This papers inference anew investment decisions or guidelines based on the minimum error percentageobtained through the mentioned performance measures. The future forecast ofeach index for next few days also highlighted in this paper.
It is hoped thatmore innovative approaches will be conducted to bring the hidden informationabout stock market. FutureworkThe work performed inthis research report comprising of two essential forecasting methodologieshowever for thefuture work scope, there is still big vacant room for testing and refining the projectedmodel by evaluating the model over other companies listed on Pakistan StockExchange there are several other variations of themodels that could be implemented to calculate the accuracy of the forecastingmodels.Other implementationsthat could provide significantly good results are:· Evaluating and Implementing the Decision tree andSupport Vector Machine (SVM) methodology to forecast the listed scripts on PSX.· Including the impact of macroeconomic indicators onthe selected sector of Pakistan Stock Exchange.
· Artificial Neural Networks, association rules andgenetic algorithms are also distinguished parameters that signify a richarea/scope for future investigations.· Using several other technical indicators inforecasting the stock scripts as well as including the fundamental analysis ofthe script as well.· Research what technical/macroeconomic indicators havebeen shown to yield good predictive qualities. Use these as inputs toclassifiers.· There are several other technical classifiers availablewhich could produce interesting results if implemented on the same data sets.· Investigate the classification rule and itscorrelation with the market volatility to perform the training data sets forforecasting.