The problem of predicting profitability is exceptionally relevant for investors and company owners. This paper examines the factors affecting firm performance and tests and compares various methods based on linear and non-linear dependencies between variables for predicting firm performance. In this study, the methods include random effects regression, individual machine learning algorithms with optimizers (DNN, LSTM, and Random Forest), and advanced machine learning methods consisting of sets of algorithms (portfolios and ensembles). The training sample includes 551 retail-oriented companies and data for 2017–2019 (panel data, 1653 observations). The test sample contains data for these companies for 2020. This study combines two approaches (stages): an econometric analysis of the influence of factors on the company’s profitability and machine learning methods to predict the company’s profitability. To compare forecasting methods, we used parametric and non-parametric predictive measures and ANOVA. The paper shows that previous profitability has a strong positive impact on a firm’s performance. We also find a non-linear positive effect of sales growth and web traffic on firm profitability. These variables significantly improve the prediction accuracy. Regression is inferior in forecast accuracy to machine learning methods. Advanced methods (portfolios and ensembles) demonstrate better and more steady results compared with individual machine learning methods.
One of the key tools in an organization’s performance management is the goal tree, which is used for solving both direct and inverse problems. This research deals with goal setting based on a model of the future by presenting the goal and subgoal in the form of concrete quantitative and qualitative characteristics and stepwise formation of factors. A stepwise solution to a factor generation problem is considered on the basis of mathematical symmetry. This paper displays an algorithm for solving hierarchical inverse problems with constraints, which is based on recursively traversing the vertices that constitute the separate characteristics. Iterative methods, modified for the case of nonlinear models and the calculation of constraints, were used to generate solutions to the subproblems. To realize the algorithm, the object-oriented architecture, which simplifies the creation and modification of software, was elaborated. Computational experiments with five types of models were conducted, and the solution to a problem related to fast-food restaurant profit generation was reviewed. The metrics of remoteness from set values and t-statistics were calculated for the purpose of testing the received results, and solutions to the subproblems, with the help of a mathematical package using optimization models and a method of inverse calculations, were also provided. The results of computational experiments speak to the compliance of the received results with set constraints and the solution of separate subproblems with the usage of the mathematical package. The cases with the highest solution accuracy reached are specified.
31 January 2020 in EUREKA: Physics and Engineering
The solution of inverse problems is considered taking into account the restrictions using inverse calculations. An algorithm is proposed for solving the inverse problem, taking into account restrictions while minimizing the sum of the absolute values of the changes in the arguments. The problem of determining the increments of the function arguments is presented as a linear programming problem. The algorithm includes solving the inverse problem with the help of inverse calculations while minimizing the sum of the absolute changes in the arguments, checking the correspondence of the obtained arguments to the given restrictions, adjusting the value of the argument if it goes beyond the limits of acceptable values, and changing the varied arguments to achieve the given value of the resulting indicator. The solution of two problems with the additive and mixed dependence between the arguments of the function is considered. It is shown that the solutions obtained in this case are consistent with the result of using an iterative procedure based on changing the resulting value to a small value until a given result is achieved, and the results are compared with solving problems using the MathCad mathematical package. The advantage of the algorithm is a smaller number of iterations compared to the known method, as well as the absence of the need to use coefficients of relative importance. The presented results can be used in management decision support systems
The digitalisation of the economy and the global coronavirus pandemic transfer competition for consumers into a virtual Internet environment, in which the enterprise success depends on its digital capital, which includes an Internet communication system with potential consumers. The article reveals the regularities and trends in the development of the enterprises digital capital focused on the retail market for the period 2017–2021. Economic and variance analysis of digital capital indicators based on the enterprises formed samples has been carried out. It has been noted that the leaders in the demand for websites are the trade and services industries. The global coronavirus pandemic in 2020–2021 became a powerful growth driver, with the major effect on enterprise website promotion is observed with a one year lag (in 2021). It has been found that the most industries enterprises are practically not engaged in the paid traffic development for their sites. Only real estate agencies actively interact with advertising Internet services. It has been shown that the enterprises digital capital development constantly requires new impulses or growth drivers and enterprises in most studied industries do not use the full opportunities range for increasing digital capital.
This paper introduces novel inverse optimization algorithms (RC and DC) for neural network training in stock price forecasting in an attempt to overcome the traditional gradient descent limitation of local minima convergence. The key novelty is a stochastic algorithm for inverse problems adapted to neural network training, where target function values decrease iteratively through selective weight modification. Experimental analysis used closing price data from 40 Russian companies, comparing traditional activation functions (linear, sigmoid, tanh) with specialized functions (sincos, cloglogm, mish) across perceptrons and single-hidden-layer networks. Key findings show the superiority of the DC method for single-layer networks, while RC proves most effective for hidden-layer networks. The linear activation function with the RC algorithm delivered optimal results in most experiments, challenging conventional nonlinear activation preferences. The optimal architecture, namely, a single hidden layer with two neurons, achieved the best prediction accuracy in 70% of cases. The research confirms that inverse optimization algorithms can provide higher training efficiency than classical gradient methods, offering practical improvements for financial forecasting.