With the computational advances over the past few decades, Bayesian analysis approaches are starting to be fully appreciated. Forecasting and time series also have Bayesian approaches and techniques, but most people are unfamiliar with them due to the immense popularity of Exponential Smoothing and autoregressive integrated moving average (ARIMA) classes of models.
However, Bayesian modeling and time series analysis have a lot in common! Both are based on using historical information to help inform future modeling and decisions. Using past information is key to any time series analysis because the data typically evolves over time in a correlated way. Bayesian techniques rely on new data updating their models from previous instances for better estimates of posterior distributions.
This talk will briefly introduce the differences between classical frequentist approaches of statistics to their Bayesian counterparts as well as the difference between time series data made for forecasting compared to traditional cross-sectional data. From there, it will compare the classical Exponential Smoothing and ARIMA class models of time series to Bayesian models with autoregressive components. Comparing the results of these models across the same data set allows the audience to see the potential benefits and disadvantages of using each of the techniques.
This talk aims to allow people to update their own skill set in forecasting with these potentially Bayesian techniques.
At the end, the talk explores the technique of model ensembling in a time series context. From these ensembles, the benefits of all types of models are potentially blended together. These models and their respective outputs will be displayed in R.
Speaker: Aric LaBarr (NC University)
www.aicamp.ai/...
28 сен 2024
