C52 - Model Evaluation, Validation, and Selection
-
-
Combining Large Numbers of Density Predictions with Bayesian Predictive Synthesis
I show how to combine large numbers of forecasts using several approaches within the framework of a Bayesian predictive synthesis. I find techniques that choose and combine a handful of forecasts, known as global-local shrinkage priors, perform best. -
Forecasting Banks’ Corporate Loan Losses Under Stress: A New Corporate Default Model
We present a new corporate default model, one of the building blocks of the Bank of Canada’s bank stress-testing infrastructure. The model is used to forecast corporate loan losses of the Canadian banking sector under stress. -
Calculating Effective Degrees of Freedom for Forecast Combinations and Ensemble Models
This paper derives a calculation for the effective degrees of freedom of a forecast combination under a set of general conditions for linear models. Computing effective degrees of freedom shows that the complexity cost of a forecast combination is driven by the parameters in the weighting scheme and the weighted average of parameters in the auxiliary models. -
Nowcasting Canadian GDP with Density Combinations
We present a tool for creating density nowcasts for Canadian real GDP growth. We demonstrate that the combined densities are a reliable and accurate tool for assessing the state of the economy and risks to the outlook. -
Estimating Large-Dimensional Connectedness Tables: The Great Moderation Through the Lens of Sectoral Spillovers
Understanding the size of sectoral links is crucial to predicting the impact of a crisis on the whole economy. We show that statistical learning techniques substantially outperform traditional estimation techniques when measuring large networks of these links. -
Survival Analysis of Bank Note Circulation: Fitness, Network Structure and Machine Learning
Using the Bank of Canada's Currency Information Management Strategy, we analyze the network structure traced by a bank note’s travel in circulation and find that the denomination of the bank note is important in our potential understanding of the demand and use of cash. -
The Trend Unemployment Rate in Canada: Searching for the Unobservable
In this paper, we assess several methods that have been used to measure the Canadian trend unemployment rate (TUR). We also consider improvements and extensions to some existing methods. -
Characterizing the Canadian Financial Cycle with Frequency Filtering Approaches
In this note, I use two multivariate frequency filtering approaches to characterize the Canadian financial cycle by capturing fluctuations in the underlying variables with respect to a long-term trend. The first approach is a dynamically weighted composite, and the second is a stochastic cycle model. -
Evaluating Real GDP Growth Forecasts in the Bank of Canada Monetary Policy Report
This paper examines the quality of projections of real GDP growth taken from the Bank of Canada Monetary Policy Report (MPR) since they were first published in 1997. Over the last decade, it has become common practice among the central banking community to discuss forecast performance publicly.