Honestly, this went better and worse than expected all at the same time.
The major problem I had originally was picking a t0 starting point. I originally used a 60-day moving average, but fans of the show will know I switched to a 30-day t0 with much better accuracy. Though I don't love the GBM method that is working under the hood, the first period (pre-holiday) is actually pretty not-bad considering the prediction was created in the most unstable period of the last year. In the future, we can probably stick to a simpler exponent model rather than the more expensive GBM model used here. For more info on how it's made, check out: Developing Price Predictions.
Given how blind a shot the prediction was, I'm extremely happy to be in the ballpark, even if the ranges at the end are pretty wide.
There are 3 big screw-ups going on here:
- t0 starting point
- GBM vs exponential modeling
- Mis-predicting discontinuity
Though the first two were expected before starting, I was a little surprised at the December performance in PLEX. Specifically over the last week where the price dropped. I did not really expect PLEX prices to slump under the 1.15B mark, or to continue their slide so strongly after black-friday. This peak/slump mirrors last year's behavior, even if the specific timings don't line up. Also, the current weakness is strange to see, where last year we could point to ISboxer policy changes driving changes.
I tried to cook "down in December" into the model, but I expected things to stay flat at 1.2B rather than slumping under 1.15B. Some of this might be attributable to Frostline drops including PLEX/MCT's, but the yields there seem extremely low compared to daily Jita trade volume.
If you were waiting for a chance to buy in, now is the time. The IRL price of PLEX is a little squishy at the moment, but things should stay in the <1.2B region until March when we should expect news/release of Citadels. I am stealing a page out of delonewolf's book to point at growth into that release and Fanfest. I will try to cook up new predictions for Q1/Q2 in the upcoming weeks. Without a date for Citadels in stone, it will be hard to get ranges correct.