Time-series prediction: key to business decision making
Professor Arnaud Dufays is an expert in Bayesian statistics. He teaches courses on data mining and time series to students on EDHEC’s Master’s in Data Analytics & Artificial Intelligence programme. Both are key to the decision-making process. He tells us all about his course.
What classes do you teach?
Currently, I teach three courses. The “data mining” and “time series” courses are taught as part of the MSc in Data Analytics & Artificial Intelligence programme. The third, called “multivariate data analysis”, is taught at pre-master’s level. These three data analytics courses provide skills in data modelling and statistical tools for making better predictions.
What can you tell us about your field of expertise?
My field of expertise is Bayesian statistics applied to time series and mostly to financial data.
Bayesian statistics is a branch of statistics that is based on the principles of Bayesian probability theory. In Bayesian statistics, probabilities are used to represent beliefs or degrees of uncertainty about the likelihood of events. In finance, Bayesian statistics can be used to model and forecast financial data, such as stock prices, interest rates or other economic indicators. This can help financial analysts and investors to make more informed decisions about their investments and to assess the risk and potential return of different investment opportunities.
One key advantage of Bayesian statistics in finance is that it allows for the incorporation of prior knowledge or beliefs about the data into the analysis. This means that analysts can use their existing knowledge and expertise to inform their predictions, which can improve the accuracy of their forecasts. In addition, Bayesian statistics allows for the incorporation of new information as it becomes available, which can help to improve the accuracy of the forecasts over time. This makes it a valuable tool for analysts who need to make decisions in fast-changing and uncertain financial markets.
Is there a must-read in your domain of expertise for students?
Bayesian statistics is different from the frequentist paradigm standardly taught in quantitative courses. I found the following book to be very rewarding: Kendall's advanced theory of statistics, volume 2B: Bayesian inference (Vol. 2) by A. O'Hagan and J.J. Forster (2004, Arnold). It discusses in detail the difference between the two types of statistics while reviewing the major concepts of Bayesian statistics. Of course, it requires some knowledge in both paradigms to fully appreciate this advanced book.
For some bedside reading, Superforecasting: The art and science of prediction by P.E. Tetlock and D. Gardner (2016, Random House) illustrates the Bayesian thinking in a prediction context (even though it is not clearly mentioned in the book). In a nutshell, the book explores the concept of "superforecasting", or the ability to make highly accurate predictions about future events. The authors discuss the factors that contribute to successful forecasting and present a framework for improving one's predictive abilities.
Why is it important for Data Analytics & Artificial Intelligence students to master time-series prediction?
Time-series prediction is a key aspect of data analytics and artificial intelligence, as it allows organisations to make informed decisions by anticipating future events based on historical data. This can help businesses optimise their operations, identify potential problems and make accurate predictions about future trends. For students, mastering time-series prediction gives them a valuable skillset that is in high demand in today's job market. In particular, understanding time-series modelling can help students to better understand and analyse complex datasets and provide valuable insights into a wide range of real-world situations.
Can you give us a concrete example of the impact on companies?
One example of the impact of time-series prediction on companies is in the field of supply-chain management. By using time-series prediction algorithms, businesses can forecast demand for their products more accurately, which can help them to optimise their inventory levels and ensure that they always have the right amount of stock on hand to meet customer demand. This can help to reduce the risk of stockouts, which can lead to lost sales and customer dissatisfaction. It can also help to prevent overstocking, which can result in excess inventory that ties up capital and takes up valuable storage space. By using time-series prediction to optimise their inventory management, businesses reduce their costs, increase their efficiency and improve their customer service.
To give another concrete example, time-series modelling could be used to implement a targeted advertising strategy. It would involve several steps. First, the business collects data on its past advertising efforts, such as the times and locations where ads were shown, the target audience for each ad and the response rates (such as clicks, conversions, etc.) for each ad. This data would then be used to train a time-series prediction model, which could be used to forecast the likely response rates for future ads based on its characteristics (time, location, audience, etc.). The predictions from the model could help to identify the most effective times and locations to show ads and to target the ads at audiences most likely to respond to them.
What are the key concepts the students will learn in your class?
Three key concepts covered in this course are stationarity, AutoRegressive Integrated Moving Average (ARIMA) models and volatility modelling.
- Understanding the concept of stationarity is essential for anyone who wants to effectively analyse and forecast time-series data.
- The ARIMA models are based on the idea that a time series can be represented as a combination of three components: a trend, a seasonal pattern and a residual or error term. Those models can be used to forecast data with trends and seasonal patterns and can also be adapted to handle non-stationary data, which are common in many real-world situations.
- Volatility modelling is useful for assessing the uncertainty of a prediction. By incorporating time-varying variance into the ARIMA model, not only will we be able to make accurate predictions of time series, but we will also be able to better understand and quantify the uncertainty around those predictions.
What are the key skills the students will gain in your class?
Students will learn how to use Python to build time-series models and to make predictions. This implies improving their knowledge of statistical concepts such as stationarity and short-memory processes, of modelling and of coding.
What do you expect them to have mastered upon completion of your course?
At the end of this course, a student will be able to use historical data to efficiently forecast time series. Given a dataset, he/she will be able to identify and account for explanatory variables, as well as for the persistence in the mean and in the variance of the series, to improve the prediction of a specific continuous time series.
How important is it for their future career?
As more and more organisations seek to make better use of their data to drive decision-making and improve their operations, the demand for skilled professionals who can effectively analyse and forecast time-series data is likely to keep growing. This means that students who master time-series modelling will be well-positioned to take advantage of these opportunities and build successful careers in these fields. In addition, understanding time-series modelling can provide students with a valuable skillset that can be applied to a wide range of sectors, for example, finance, marketing and healthcare.
Discover the MSc in Data Analytics & Artificial Intelligence