The following pictures represent the real 5-Layer forecast model provided on the CRUDE-OIL for 3 months, with the real price that enables the post analysis 3 months later.
The original forecast provided:
The post forecast 3 months later:
Hope you are doing well.
I have added the original picture sent and also ran the same setup with the exact marked dates and months, but now with the real price, to make it easier to compare.
It is essential for me to look back and examine the performance of the forecasts. Although I am doing that regularly, I am happy to make that with you and analysis the performance of each model and the overall.
I am not expecting 100% accuracy but still want to be able to explore when the forecasts are more reliable, or alternatively, how we should better read the graphs.
Here is my analysis:
1. Being able to reproduce the exact same projection lines with NN is a good point as NN run may not be identical by nature.
2. All models show up trend until 3-Feb - OK,
3. All models show down trend from Feb-20 - OK
4. Feb-3 is the point where the long-term model starts to bent to the downward direction, other models show a peak. That may rpresent a near lower-high.
1. Dark-blue Yearly model: Peak around mid-January - OK
2. Blue Quarterly model: Good direction, Peak at 3rd part of Feb instead of early Feb - NOK (may reprsent a near lower-high that will form a the right side of a head-and-shoulder structure).
3. Green Monthly model: Good direction, bent at beginning of Feb - OK, Peak at 3rd part of Feb instead of early Feb - NOK (may reprsent a near lower-high that will form a the right side of a head-and-shoulder structure).
4. Orange Weekly model: Good directions, Peak at Feb-3 - OK, Low at Feb-10 - OK, Peak Feb-20 - OK, Peak at Mar-6 - OK, Low at mid-Mar - OK.
5. Velvet Daily model: Good directions, Peak at Feb-3 - OK, Drop at Feb-10 - OK, Peak Feb-20 - OK, Peak at Mar-6 - OK, Drop at mid-Mar - OK.
Overall directions were good, turning points dates were good, Blue and Green models were a bit off peak. This may reprent the option that there a lower high yet to be performed, which will generate a head-and-shoulder structure.