RWE npower Forecasting Challenge
[This article was first published on Q-Day, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Lately I have been involved in organising forecasting competition for undergraduate students. The obvious goal is to find someone with forecasting power who would potentially be interested in internship or graduate scheme. Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Recently, I asked a few people from quant and forecasting teams to try the task out in order to validate the data and get some opinions about the difficulty of it. What happened is that after few attempts it was clear that half-hourly data might be a bit too detailed for 2nd or 3rd year student. I don’t think I have seen a dataset with more than 500 observations during my undergrad while data had 70000.
Finally it was agreed to aggregate the data to daily granularity and we tested it again. This time I got forecasts promptly and noticed that some competition is arising internally. This has led to allowing internal applicants to attempt the task. It was agreed that main competition rating ta able will be kept separate, but all applicants including internal will have a separate one – I bet graduates will also be interested how well they did against someone who is actually working in forecasting.
Finally, I thought that it doesn’t make much effort to also allow external non-student participants. Thus we agreed that anyone can apply.
The task is quite simple:
Train on 2 years, provide forecasts for next 6 months.
Then get actuals for those 6 months, train again, forecast for next 6 months.
And do the same again.
Total 3 rounds, measure is MAPE – mean(abs(percentage error)).
I quite enjoyed attempting the task myself and having unfair advantage of being able to test on actuals outperformed everybody. However, the same model has performed even better in rounds 2 and 3 without any further adjustments.
I don’t think it would take more than an hour for somebody who knows his/her R and might be a good fun seeing yourself against results of internal RWE npower forecasting team, internal quants and most importantly graduate students! 🙂
“Please note this competition is for University students only (see terms & conditions). However if you would still be interested
in attempting the challenge please email [email protected]. You will not be eligible for the prize but your forecast will be scored and appear in our rankings.”
Note amended e-mail!
For more information visit: http://www.npowerjobs.com/graduates/forecasting-challenge-2015
Good luck in forecasting and hope to see you in high scores!
By the way this is how my first train model looked like. (Different dataset though)
To leave a comment for the author, please follow the link and comment on their blog: Q-Day.
R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.