Deep Learning Based Financial Forecasting Models

Abstract

Financial planning involves systematical forecasting and calculation of cash and financial flows into and out
of the company. Financial planning is the reconciliation of cash inflows and outflows, both in terms of amount
and time by forecasting all types of cash inflows and outflows that will occur during the company's operations.
It allows to quickly determine the solution process, make analysis, forecasts and strategic decisions. This
study aims to develop financial forecasting models using univariate deep learning methods. For this purpose
Long Short Term Memory (LSTM), Bi-directional Long Short Term Memory (Bi-LSTM) and Convolutional
Long Short Term Memory (ConvLSTM) have been used. The performance of the developed models has been
evaluated using Mean Absolute Percentage Error (MAPE). The dataset includes 464 rows and total inbound
and total outbound invoice amount data from June 22nd, 2020 to March 31st, 2022. Forecast models have
been developed for 2 different weeks (28.02.2022 – 04.03.2022 and 21.03.2022 – 25.03.2022) and 2 different
months (January 2022 and March 2022) randomly selected from the dataset. When the forecast models
developed for inbound invoice amount and outbound invoice amount are examined, it is found that
satisfactory results have not been obtained for the monthly forecasts. For the weekly forecasts, MAPE’s of
the forecast models were found to be less than 20% in general.