Abstract: Time series analysis is both a fascinating subject to study and an important set of techniques that enjoy a wide range of applications in industry, government, and academic settings. Use cases range from inventory management, capacity planning, marketing strategy design, capital budgeting, pricing, macroeconomic forecasting, and supply chain forecasting.
A common aspect to all of these applications is the use of forecasting, and time series forecasting requires time series data that is ubiquitous nowadays: weekly initial unemployment claims, product-level hourly sales, tick-level stock prices, daily term structure of interest rates, quarterly company earnings, daily number of steps taken recorded by a wearable, machine performance measurements recorded by sensors, and key performance indicators of business functions, just to name a few.
Time series data differs from cross-sectional data in that time series data has temporal dependence, which can be leveraged to forecast future values of the series. Some of the most important and commonly used data science techniques to analyze time series data and make forecast based on them are those in developed in the field of statistics and machine learning. For this reason, time series statistical and machine learning models should be included in any data scientists’ toolkit.
This workshop teaches the application of two important classes of time series statistical models (Autoregressive Integrated Moving Average Model and Vector Autoregressive Model) and an important set of neural network-based algorithms (Recurrent neural network) in time series forecasting. The attendees will learn the mathematical formulation, python implementation, the advantages, and disadvantages of when using these techniques in time series analysis. Jupyter notebooks with examples and sample codes will be provided for attendees to follow along and experiment with these techniques.
This workshop is divided into four parts, and a 15-minute break will be given between Part II and Part III.
Part I starts with the introduction to time series analysis, which includes the formulation of the time series problem, basic terminology, essential concepts, and the steps to analyze time series data.
Part II discusses the class of Autoregressive Integrated Moving Average Model (ARIMA), mathematical formulation, lag operator representation, model estimation, model diagnostics, model identification, model selection, assumption testing, statistical inference, and forecasting for both stationary series, non-stationary series, and series with seasonality.
Part III studies Vector Autoregressive (VAR) model, an important class of multivariate time series models. Similar to Part II, we will cover mathematical formulation, lag operator representation, model estimation, model diagnostics, model identification, model selection, assumption testing, statistical inference, and forecasting for both stationary series, non-stationary series, and series with seasonality.
Part IV introduces the application of recurrent neural networks to time series forecasting, covering the issues of using the basic feedforward network for modeling time series data, the various forms of recurrent neural networks, and the implementation in Keras.
The workshop concludes with a comparison of the various methods for time series analysis.
To fully appreciate the topics covered in this workshop and follow along with the examples, the attendees should have the following background:
1. Strong understanding of classical linear regression modeling
2. Working knowledge of Python
3. Basic understanding of neural network-based modeling
Bio: Jeffrey is currently the lead lecturer at UC Berkeley, School of Information. He was the Chief Data Scientist at AllianceBernstein, a global investment firm managing over $500 billions. He was responsible for building and leading the data science group, partnering with investment professionals to create investment signals using data science, and collaborating with sales and marketing teams to analyze clients. Graduated with a Ph.D. in economics from the University of Pennsylvania, he has also taught statistics, econometrics, and machine learning courses at UC Berkeley, Cornell, NYU, the University of Pennsylvania, and Virginia Tech. Previously, Jeffrey held advanced analytic positions at Silicon Valley Data Science, Charles Schwab Corporation, KPMG, and Moody’s Analytics.