1
$\begingroup$

Is machine learning a good tool to use in order to forecast inflation for the short term (next 2 - 3 years on a monthly or quarterly basis)?

I want to be able to forecast inflation for Canada for the next couple of years and I am trying to explore some more creative approaches to forecasting.

I am aware of the more traditional time series forecasting models like ARIMA and VAR, and I have made a basic ARIMA and VAR forecasting model in R, and I plan to make a VARX also. However, I was wondering if using neural networks is a good idea to try out. It seems to be a newer approach to time series forecasting and I'm not sure how it fits in an economic sense.

Would modeling CPI inflation and making forecasts for the next couple of years on a monthly (or quarterly) basis using machine learning of neural networks be a good idea and make economic sense? If not, are there other unconventional ways to forecast inflation?

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ Just my 2cents. 3 years is already a very long term forecast for economic variables. It's more or less the maximum you will find from professional forecasters and most only forecasts shorter periods than that. Also, if a central bank is truly committed to an inflation goal, say 2%, your best forecast would simply be that value for such a long term. If inflation deviates, its usually unforeseen events like wars, economic crisis etc where ne forecasting model will do well. $\endgroup$
    – Alex
    Oct 22, 2022 at 7:13
  • $\begingroup$ bankofengland.co.uk/-/media/boe/files/events/2020/november/… $\endgroup$
    – EB3112
    Oct 22, 2022 at 9:31
  • 1
    $\begingroup$ The question here is which neural network you use. Architecture, inputs, and implementation are extremely important. No doubt, one model in space of possible models is extremely effective. Unclear which one that would be though. $\endgroup$
    – RegressForward
    Oct 22, 2022 at 14:15

1 Answer 1

Reset to default
1
$\begingroup$

I am far from being an expert, but would say it is not the model but the data that can help you gain forecast accuracy in this case. Modern machine learning applications in macroeconomic forecasting exploit new sources of data and there lies their relative success. If you just take your 3 or 4 favorite macroeconomic indicators (GDP, unemployment, inflation, ...), build a VAR model and forecast from it, that is probably close to the best forecast you can get from that data. I doubt a fancy neural network trained on the same dataset would beat your VAR. But if you have some additional, high-frequency data (perhaps Google searches of relevant keywords), then a fancy neural network might help a bit more.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thank you for your response, I was thinking about incorporating google trends in relation to inflation expectations, but do you have any advice on high-frequency data that would be worth using to make a neural network forecast? $\endgroup$
    – eddie
    Oct 22, 2022 at 16:00
  • $\begingroup$ @eddie, sorry, I do not. I am not working on macroeconomics. $\endgroup$
    – Richard Hardy
    Oct 22, 2022 at 16:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.