Try it: getstatmate.com
And the two most severe bugs in GetStatMate's recent work are they didn't crash anything.
They simply returned incorrect answers.
- ADF falls over small samples. Statsmodels used 9 lags as default on 22 years of annual data.
On differenced inflation, it gave a positive t-stat and p = 0.997, and nothing errored. When I did it by SIC (up to the cube root of the sample size), it fixed itself.
Moral of the story: "it ran" isn't very important in a statistical script.
Test on instances where you've previously determined the result.
- Verifying indicator codes. A review of all World Bank codes against the API identified roughly 20 incorrectly labelled indicators, 6 deleted indicators, and roughly 15 indicators with no recent Nigerian data. New too: a unit root table, one-click transforms and CBN data imported directly from the endpoints behind the CBN site.
ARDL is next.
Try GetStatMate
Top comments (1)
tr.ee/dev-to