Economic analysis of risky projects by ANNs
Introduction
The artificial neural network (ANN) model development is a technique heavily researched and used in applications for engineering and scientific fields for various purposes ranging from control systems to artificial intelligence. Its generalization powers have not only received admiration from the engineering and scientific fields, but also in recent years, the finance researchers and practitioners are taking interest in the application of ANN [28]. Risk assessment application is one of the areas that is heavily researched in the financial field.
In dictionary definition terms ‘risk’ means: “hazard, chance of bad consequences, loss, exposure to chance of injury or loss” (Concise Oxford Dictionary). Such definitions illustrate one problem with the term ‘risk’—its ambiguous use as a synonym of probability or chance in relation to an event or outcome, the nature of an outcome, or its cause. Dowie [6] argues that the term ‘risk’ is an obstacle to improved decision and policy making. Its multiple and ambiguous usages persistently jeopardize the separation of the tasks of identifying and evaluating relevant evidence on the one hand, and eliciting and processing necessary value judgements on the other. The term ‘risk’ contaminates all discussions of probability because of the implicit value judgement/s that the term always brings with it, just as it contaminates all discussions of value assessment because of the implicit probability judgement/s that it contains [6], [32].
Risk management is a critical part of project management as ‘unmanaged or unmitigated risks are one of the primary causes of project failure’ [29]. While numerous papers have been written on the subject of risk management, little current information exists on the actual use of risk management in practice [22].
The management of risk in projects is currently one of the main topics of interest for researchers and practitioners working in the area of project management [27]. The concept of risk most widely used in project evaluation is the variability of return, which is measured by variance (or standard deviation). The use of variance as a measure of risk implies that deviations below the expected value are regarded in the same way as deviations above the expected value [24]. Most project managers are risk averse, in the sense of utility theory. Hence, they would most likely not like the variation in project costs that would occur if projects were accepted as investments based on their expected cost (or profit) [15]. Identification of project risk is, in the main, achieved by analysing the financial data through the measurement of its ‘sensitivity’ to variations and by identifying a project’s payback period [19].
According to McMillan [23], cost estimation is difficult in certain industries, often leading to considerable cost overruns. The explanations are that there is often large uncertainty—often related to new technology—and that the uniqueness of the project, limits the learning process. One might expect that the cost overruns have the same probability as completing the project below the cost estimate. However, observations clearly indicate an overrepresentation of cost overruns. This is due to two types of selection bias : (1) project selection; it is typically the projects with the most optimistic internal cost estimates that are being pursued by an investor, and (2) tender selection; competition sees to it that tenders with pessimistic and realistic cost estimates are ruled out [7].
In this paper, we present an artificial neural network model for economic analysis of risky projects. MLP and RBF are used for approximating the risky projects. The performance of Radial Basis Function was compared with multilayer perceptrons.
Section snippets
Artificial neural networks (ANN)
Artificial neural networks represent a connection of simple processing elements capable of processing information in response to external inputs [1], [8], [11]. The most common architecture of ANN’s is in the form of layers of neurons [10]. Fig. 1 shows a neuron structure.Fig. 2 shows the general structure of the one-hidden-layer of ANN. The ANN’s input layer with some neurons represents the previous data, say period t − p to period t − 1, which are connected
Analytical analysis of risky projects
In economic analysis, it is often assumed that all factors are deterministic in nature. In reality, factors such as initial investment, revenue, rate of return and investment life have stochastic properties [3]. In some cases where the stochastic nature of a factor is recognized, some flaws may still exist because of the simplifying approach that is used in the analysis. Two of the common, but wrongful, practices in analytical approaches involve representing a distribution simply by its mean
Implementation and results
Function approximation of neural networks were applied to the cost estimation of risky projects. The training data were obtained from Fig. 3, Fig. 4 where X and Y are normally distributed with (μ, σ2) known: X ∼ normal (1, 0.5) and Y ∼ normal (2, 1).
Using the X and Y variables, the total cost of project CT is obtained from Eq. (6). To obtain the variance (Var[CT]) and expected cost value (E[CT]) Eqs. (11), (12), (13) were used and finally they were substituted into Eqs. (14), (7). The training data in
Conclusion
In this study, two main ANN model RBF and MLP were applied to model the risky economical projects and performance results were compared. Analysis of the neural network outputs proved that more predictive capability can be achieved by MLP. The proposed network can be applied to any risky economical project analysis if the project inputs are normalized with a Gaussian distribution function.
Acknowledgement
Special thanks to Mutlu Avci, from Yildiz Technical University Department of Electronics and Communications, for his help.
References (34)
- et al.
Neural network as a simulation metamodel in economic analysis of risky projects
European Journal of Operational Research
(1998) - et al.
Investment cost estimates and investment decisions
Energy Policy
(2002) - et al.
Improving project cost estimation by taking into account managerial flexibility
European Journal of Operational Research
(2000) - et al.
An intelligent sales forecasting system through integration of artificial neural network and fuzzy neural network
Computers in Industry
(1998) - et al.
A new pragmatic approach to capital investment appraisal: The financial appraisal profile (FAP) model
International Journal of Production Economics
(1998) - et al.
Fault detection and diagnosis using statistical control charts and artificial neural networks
Artificial Intelligence in Engineering
(1998) - et al.
A neural network model with bounded-weights for pattern classification
Computers and Operations Research
(2004) - et al.
Project risk management in the Queensland engineering construction industry: A survey
International Journal of Project Management
(2004) - et al.
Use and benefits of tools for project risk management
International Journal of Project Management
(2001) - et al.
Transforming project risk management into project uncertainty management
International Journal of Project Management
(2003)
Expert Systems Applications in Engineering and Manufacturing
Stochastic Rate of Return Analysis on a Microcomputer, Softcover Software
Generalization properties of radial basis functions
Multivariable functional interpolating and adaptive networks
Complex Systems
Against risk
Risk Decision and Policy
Neural Networks: Algorithms, Applications, and Programming Techniques
A constructive method for multivariate function approximation by multilayer perceptrons
IEEE Transactions on Neural Networks
Cited by (11)
Artificial neural networks in business: Two decades of research
2016, Applied Soft Computing JournalA new methodology for multi-echelon inventory management in stochastic and neuro-fuzzy environments
2010, International Journal of Production EconomicsCitation Excerpt :Model-G1 and G2, that Model-G2 has a neuro-fuzzy module to compute demand and lead times, are simulated through an artificial neural network (ANN). ANNs are heavily used in the engineering and scientific fields to model systems ranging from control systems to artificial intelligence (Taskin and Guneri, 2006; Guneri and Taskin Gumus, 2007). ANNs are networks of simple processing elements capable of processing information in response to external inputs (Hecht-Nielsen, 1989; Freeman and Skapura, 1991; Badiru, 1992; Haykin, 1999).
Supply chain network design using an integrated neuro-fuzzy and MILP approach: A comparative design study
2009, Expert Systems with ApplicationsA multi-echelon inventory management framework for stochastic and fuzzy supply chains
2009, Expert Systems with ApplicationsShort-term electric load forecasting based on LS-SVMs
2012, Advances in Intelligent and Soft Computing