Machine Learning for Financial Risk Management with Python

Chapter 2. Introduction to Time Series Modeling

Market behavior is examined using large amounts of past data, such as high-frequency bid-ask quotes of currencies or stock prices. It is the abundance of data that makes possible the empirical study of the market. Although it is not possible to run controlled experiments, it is possible to extensively test on historical data.

Sergio Focardi (1997)

Some models account better for some phenomena; certain approaches capture the characteristics of an event in a solid way. Time series modeling is a good example of this because the vast majority of financial data has a time dimension, which makes time series applications a necessary tool for finance. In simple terms, the ordering of the data and its correlation is important.

This chapter of the book will discuss classical time series models and compare the performance of these models. Deep learning–based time series analysis will be introduced in Chapter 3; this is an entirely different approach in terms of data preparation and model structure. The classical models include the moving average (MA), autoregressive (AR), and autoregressive integrated moving average (ARIMA) models. What is common across these models is the information carried by the historical observations. If these historical observations are obtained from error terms, we refer to this as a moving average; if these observations come out of time series itself, it is called autoregressive. The other model, ARIMA, is an extension ...