Michael Creel
michael.creel @ uab.cat
Tel. +34 93 581 1696
Office B3-1104
Departament d'Economia i d'Història Econòmica
Michael
Creel
michael.creel @ uab.cat
Tel. +34 93 581 1696
Office
B3-1104
Departament d'Economia i d'Història Econòmica
Econometrics notes Graduate level notes for econometrics, with numerous examples using the Julia language embedded as links.
Fall 2023: Constructing Efficient Simulated Moments Using Temporal Convolutional Networks (with Jonathan Chassot). This paper continues work with simulated moments estimation using neural nets. In this case, the net, which uses the TCN architecture, takes the entire sample as the input, instead of statistics. Thus, no information is lost due to selection of statistics. We show by example that MSM estimation using the output of the net as the statistics for indirect inference can equal or beat maximum likelihood estimation, in terms of RMSE, for several test models, when the sample size is not too large. The methods are applied to a jump diffusion model for S&P500 data. One result is that measurement error is clearly present.
29 Sept. 2021: “Inference Using Simulated Neural Moments”. This short paper looks at how estimators that simulate from a model and base estimation on statistics (moments) can benefit from filtering the statistics through a trained neural net to improve reliability of inferences. This works pretty well, I am pleased to say. There are 4 example models, including a small DSGE model, and there is an empirical application of a jump-diffusion model for S&P 500 returns. See the code archive https://github.com/mcreel/SNM. This paper continues the project started in my paper “Neural Nets for Indirect Inference”.