A neural network is an advanced computer artificial intelligence system with the ability to capture, analyze, and represent complex relationships in data that can be used for decision-making. The system can compare current stock-trading patterns using previous situations, analyze all available indicators, and ultimately learn the output that works, as the system digests more data. The relevance and advantage of neural network rests is on its ability to process large amount of data, determine both their linear and non-linear connections based on the trading patterns, and its ability to make complex analysis that cannot be completed by human analysts at the same time (Cheng & McClain 1997). The neural networks use extensively parallel-distributed processors that constitute simple processor unit with natural ability to store previous experiential knowledge and making it available to aid future analysis. In this case, the artificial neural networks (ANN) is a model used to compute non-linear data and has the adaptive system that varies its structure to find successive patterns depending on the external or internal data that runs through the network during processing. This ability of ANN allows management teams to make accurate forecasting of future market performance and to make correct investment decisions that will enable optimum utilization of resources.
Statement of the Research Problem
Accurate predictions of future market performance in the stock markets, accurate estimation of inflation rates and forecasting financial markets necessitates a dependable and simple model. This has been a challenge to managers for a long time, owing to the large volumes of data and information that need to be processed to understand the relationships to aid accurate forecasting. To solve this problem, neural networks qualifies to be a preferred tool that can be used for several predictive data mining applications, as a result of their ability and flexibility in analyzing complex data. The system can accurately forecast consumer demand to plan for production, predict possible response to marketing activities in terms of volumes, predicting credit risks, and detect fraudulent transactions and claims in insurance database among others. This accuracy is derived from its ability to perform tasks that linear programs cannot do, its parallel system that allows continuous processing even if element of the neural network fails and the ability of the neural network to learn, not to require reprogramming when a new challenge is met.
Here you can get a price quote:
Rationale of the Study
This paper looks at the neural networks as a model that can be used to aid managers in making accurate investment decisions. Previously, processing large quantities of data at the same time by people or preprogrammed systems has been an exerting substantial challenge. The challenge has been on the promptness and accuracy since the preprogrammed systems of conventional computers can only use a set of instructions to handle the predetermined cases, but cannot solve problems that are not already understood. Neural networks, on the other hand, use highly interconnected processing elements (the neurons) that work in a parallel manner just like the human brain to solve a specific problem, thus increasing accuracy (Amaral & Passador 2012). Therefore, it is important that managers fully understand how neural networks operate for them to apply the processes for the accurate and efficient decision-making.
The Neural networks under focus in this study for predictive applications is the radial basis function (RBF) and multilayer perceptron (MLP) networks since the model-predicted results can be compared against the values of target variables.
Statement of the Research Questions
The research questions that will be used to achieve the objectives of the study are (1) can neural networks performs accurate forecasting of nonlinear elements? and (2) between the multilayer perceptron (MLP) and radial basis function (RBF) networks, which model performs better? These two questions will adequately address study objectives.
Objectives of the Research
The objectives of the study are
- To investigate whether neural networks can aid managers in performing accurate forecasting of nonlinear elements necessary for the decision-making.
- To determine which model between the radial basis function (RBF) and multilayer perceptron (MLP) networks performs better in analysis.
- To understand how neutral networks operate and the interpretation of results for the investment decision-making.
Limitations of the Study
This study was limited to the radial basis function (RBF) and multilayer perceptron (MLP) networks perform. There was no comparison with other neural networks to identify the best one.
This chapter will analyze secondary information related to the neural networks and investment decisions. This will be done through understanding the relevant terminologies that explain neural networks and review of the articles related to the how neural networks models can be used to accurately forecast future market performance in order to improve investment decision making process in the financial institutions.
this is an artificial neuron, a device comprising several inputs, processing point as the hidden layer but a single output. It has two modes of operation – the using mode and the training mode. The training mode is where the neuron can be trained to fire for specific input patterns, while the using mode is where the associated output of a taught input pattern detected at the input becomes the current output. This means that in case the input pattern does not correspond to the taught list of input patterns, the network uses its firing rule to determine whether to fire or not (Mittal 2012).
this refers to the artificial neural network, which is the inter–connections between the neurons found in the different layers of each system that facilitate communication between input elements to the processing hidden layer and to the outputs. The basic neural network has three layers. The first layer of the function involves input neurons, which relay data through synapses to the next layer of neurons (hidden layer) and further to next synapses to the last level of output neurons (Mittal 2012). More complex ANN usually has more layers of neurons and the synapses store parameters called known as ‘weights’ that are used to manipulate the data in the calculations.
Trained Neural Networks
these are neural networks with extraordinary ability to generate meaning from complex and imprecise data and be able to extract the patterns as well as detect trends which are complex to be identified either humans or the conventional computer programs. A trained neural network is considered an expert that predicts market behavior through adaptive learning where the ANN learns to perform subsequent tasks based on data fed during training or initial experiences, self-organization where ANN creates its own representation of the information fed during learning time and real time operation (Mittal 2012).
Unsupervised and Supervised Learning
supervised learning is a case in which the network is trained by matching the input to the output patterns. This input-output matching can be given by the external teacher or by the system that has the neural network. On the other hand, unsupervised learning refers to self-organization where an output unit is trained to respond to a collection of pattern within the input (Mittal 2012). The system is expected to discover statistically relevant features of the input population.
According to the study done by Amaral and Passador (2012), the estimation of inflation rates is fundamental for managers since the investment decisions are closely connected to them. Nevertheless, the behavior of inflation rates is usually nonlinear, thus making it difficult to be predictable. The study by the two researchers that involved three basic models of Multi-Layer Perceptron ANN appropriated that the ANN model was adequate in measuring the phenomena of inflation through the polynomial functions that is capable in handling nonlinear phenomena. This article successfully evaluated the effectiveness of ANN in forecasting inflation, especially in small organizations through formal statistical analysis. The paper concluded that certain models of ANN could reasonably forecast inflation rates in the short run.
In addition, Sugathan and Baid (2013) found that the Private equity (PE) fund managers persistently faced a challenge of identifying viable firms to invest in that could optimize their investment decisions. Such funds face greater risk than the traditional investments, thus require more due diligence. Analyses of the factors affecting investment decisions, especially for PE funds, indicated that only two models could be employed to determine the optimum investment firm from a number of likely firms. These models are the neural-network-based optimal investment model and the risk-weighted fuzzy optimal investment model.
An article published by Maknickiene and Maknickas in 2013 also noted that the use of artificial intelligence systems for forecasting financial markets needed a simple and reliable model that would safeguard profitable growth. The two presented a model that combined Evolino neural networks with that of orthogonal data inputs, and the Delphi expert evaluation method for analysis to aid the decision-making process relating to investment portfolio. The study indicated that the model was reliable and accurately predicted financial markets performance for the accurate decision -making. This means that intelligence systems are viable.
According to An-Sing and Leung (2005), the neural networks have arose from being secret instrument used for academic research to popular tools used by investors, investment advisors, portfolio managers, and auditors in making vital financial decisions. The article notes that better understanding of how the networks perform and of their limitations is significant to the practitioners in analyzing real-world problems. This study compared the performance of models using the multi-layered feed forward neural network (MLFN) and the general regression neural network (GRNN). Results from this study pointed that the choice of architectural design might contribute to the success in neural network forecasting. Equally, the market timing tests showed that both MLFN and GRNN models were economically significant in forecasting the correlation of exchange rate, thus ensuring that neural networks were viable tools of forecasting.
Cheng and McClain (1997) additionally asserted that ANN systems had the ability to process large amounts of data than humans. In this case, a properly developed ANN investment system learns over a period of time in order to accurately predict mort by matching past predictions to actual results and further adjusting the approach to limit the differences in the future. The article notes that despite the fact that ANN systems are still in the initial stages of development, they already outdo traditional investment forecasting techniques since no other tool can receive and accurately process such huge amounts of nonlinear data, be able to discover the correlation within the data, and make accurate future forecasting.
Brown (2006) in his article also points out that neural networks, which enhance artificial intelligence capabilities, potentially improve the investment decision-making process. The neural networks integrate fuzzy logic to function as human brain determining relationships among data sets.
Finally, Roy (2005) demonstrates using MATLAB’s Perception model the acceptable and unacceptable trade in publicly traded stock. He notes that Perceptron models have been successfully used in the past. Using the readily available financial data such as current ratio, quick ratio and the sales/asset turnover among others, the initial results using the neural network model showed 90 percent accuracy in prediction and was necessary for investment decision process.
Summary, Comments, and Criticisms
All evidences indicate that ANN systems aid managers in accurately forecasting market performances must faster because of its ability to process huge data of nonlinear nature and be able to learn and store the analysis for future use. Therefore, it is observed that managers adopt the models to enhance accurate investment decisions. However, the development of these models must be accurate to avoid error in processing data, especially, in the case of trained neural networks.
To efficiently understand the accuracy in predicting market performance and to determine which model between the multilayer perceptron (MLP) and radial basis function (RBF) networks performs better in analysis, the researcher will run a test to predict the level of loan default rate by customers in order to guide formation of lending policies and to minimize the risk of loan defaulting.
Sample Data and Period
The study will use secondary data that will be collected through data mining of the previous loan issue records available in the bank. Using multilayer perceptron in forecasting the future default rate, the sample used will involve 499 cases assigned to the training sample and 201 to the holdout sample, while 150 cases excluded from the analysis are the prospective customers.
The multilayer perceptron of ANN indicates the information useful for making sure that the specifications used in the network are correct. The number of units in the input layer is the number of covariates added to the total number of factor levels. A separate output unit is generated for each classification defined as previously defaulted for the loans to be used in prediction of the rate of future loan defaulting.
Upon running the test, the model summary indicates information regarding the results of training as well as applying the final network to the holdout sample. This error function is what the network tries to reduce during training.
Using radial basis function (RBF) networks to do the forecasting, the following model summary is generated. The classification table indicates that 80.4% of the training category will not default loan payment. This gives 81.6% correct prediction for the training category and 80.0% correct prediction on default rate for testing category. This means that multilayer perceptron gives a more accurate prediction at 84.8%, hence answering the research question.
Summary of Findings, Suggestions, and Recommendations
Artificial neural networks (ANN) is a model used to compute non-linear data and has the adaptive system that varies its structure to find successive patterns depending on the external or internal data that runs through the network during processing. This ability of ANN allows management teams to make accurate forecasting of future market performance and to make correct investment decisions that will enable optimum utilization of resources. In this study to forecast the rate of future loan default by customers, it was found that the multiplier networks gave the best forecasting result when compared to the radial basis function. Based on the findings, it is strongly suggested and recommended that investors, portfolio managers, auditors, and investment advisors to use ANN in predicting future performance to aid the accurate decision-making.
This paper looked at Neural Networks and investment decisions and found that neural networks had the ability to process huge amounts of data and accurately forecast the future better than humans. Based on the literature revew and findings of the study, the multiplier networks works better in the forecasting.
Huge First Discount
Mobile Friendly Website
Friendly Support 24/7