Design and development of a forecasting tool for the identification of new target markets by open time-series data and deep learning methods

Abstract

Globalization and rapid advancements in the IT sector brought new challenges and intensified competition between companies, strongly highlighting the demand for Business Intelligence and Analytics in decision-making and strategy development planning. Motivated by the analysis and forecasting capabilities offered by time-series data and its limited exploitation in marketing literature, this paper introduces a multidisciplinary framework, called MULTIFOR, which aims to synthesize and deliver profound knowledge about the macroeconomic environment and the attractiveness of new target markets. MULTIFOR main objective is to fully support interested parties (i.e., industries, SMEs, scholars, local and national authorities, etc.) in their decision-making and strategy planning tasks, by providing accurate forecasts and recommendations. The proposed framework along with its Web service offers a unique solution since it encapsulates: marketing fundamentals (PESTEL analysis), open data (time-series data from open Web databases) and deep learning methods (LSTM networks) for time-series analysis and forecasting. MULTIFOR has been tested and validated through a use case on elevator and escalator (E&E) industry, which is a strong industrial sector which has important impact worldwide, with studying its scope in European markets. The research findings revealed that LSTM networks, which according to the existing literature are superior to other forecasting models, when combined with macroeconomic theory can achieve greater forecasting accuracy in identifying new international markets. In the case of MULTIFOR, the improvement of LSTM networks performance was achieved through the selection of appropriate indicators and the pre-processing of time-series data exploiting PESTEL analysis. In particular, MULTIFOR achieved approximately 70% of the 900 time-series reduction of errors in the forecasting process in time-series derived from the PESTEL analysis for European countries. Moreover, the exploitation of MULTIFOR in the E&E industry, revealed the countries of northern Europe as the most attractive markets, of which Sweden and The Netherlands stand out.