Bingqing-econ / aiyagari

Bewley-Huggett-Aiyagari model

We compare the elapsed time for solving the standard Aiyagari model by each code written in Matlab, Python (with or without numba), or Julia.

We run the scripts on a machine with Intel Core i7-4790K (4.0Ghz). To measure the time, we use hyperfine.

For Matlab (R2020a),

hyperfine 'matlab -batch "aiyagari"'

For Python (3.7.4 with Numba 0.45.1),

hyperfine 'python aiyagari.py'

For Julia (1.2.0),

hyperfine 'julia aiyagari.jl'

Results

	Julia	Python	Python-Numba	Matlab
Time (mean ± σ):	37.274 s ± 0.239 s		24.002 s ± 0.482 s	31.899 s ± 0.260 s
Range (min … max):	36.961 s … 37.758 s		23.551 s … 24.950 s	31.535 s … 32.215 s
Runs	10		10	10

References

https://www.sas.upenn.edu/~jesusfv/Update_March_23_2018.pdf

https://web.stanford.edu/~maliars/Files/CEPR-DP13210.pdf

Acknowledgement

I thank @yirwk and @yukiyao for improving the python code.