The Data Driven Model for Earnings Forecasting

Aske Buemann

Student thesis: Master thesis


Earnings forecasting is a central topic for many financial statement users in order to predict the future performance of a company. These users can generally be grouped in 4 categories. Firstly, investors are dependent on the information to identify the correct price of a stock to implement in their trading strategies. Secondly, analysts generate and use the forecasts for providing recommendations. Thirdly, venture capitalists use future earnings to value non-listed companies. Finally, forecasting earnings is imperative for any managerial budgeting and resource allocation process.
Despite its unneglectable importance, the area of research has been dominated by the same two types of models for the past decades. Only recently, the time series and analyst forecasts have been challenged, but no accurate contender has been identified yet. One of the main challenges in this regard is the requirement for a generally applicable model in order to meet the variety of needs within the earnings forecasting community. Therefore, the ideal model is able to apply to listed and non-listed companies across a range of sizes and industries.
The thesis proposes an alternative method for earnings forecasting that it designates the data driven model. The model represents a way of combining the most essential advantages from the time series and analyst methodologies, whereby it integrates more information than the time series models and is less biased than the analyst forecasts. Using a larger and broader dataset than seen in previous literature, the thesis proves the superiority of the data driven model over both the time series models and analyst forecasts. In this way, the thesis contributes to the earnings forecasting literature with a new model and the underlying reasoning that utilizing financial information from comparable companies produces better forecasts than solely relying on past earnings of the same company.
The superiority of the alternative forecasting methodology has several implications for the financial statement users. Using the generally applicable data driven approach, the venture capital community is now able to generate reliable forecasts for non-listed firms that are superior to the previously used time series forecasts. At the same time, investors are able to implement better and almost costless forecasts of listed companies’ earnings in their trading strategies, while managers can ease the earnings estimation in their budgeting and resource allocation processes.

EducationsMSc in Finance and Investments, (Graduate Programme) Final Thesis
Publication date2017
Number of pages109
SupervisorsJeppe Christoffersen