SPY needs to show a little more strength to get to xFC.
Aggregation just dumps all the 3x Bulls together. You can do that with natural logs and they are all playing the same days. The numbers have to be divided by 8 to get the “real” number of days. For example, xM3 has Days = 8 but it only happened once. Note the bullish states, xFF, xCF, xCC, and xFC happen the most. That is normal, the two year period is tilted to getting more than its usual share of these signals.
xCC and xFC are shown above. xCC is clearly favorable to be long, xFC is less clearly favorable to be flat. The numbers for xFC tend to negative: negative CC, negative HLd, however CO is OK. The algorithm will not buy there unless TF is active which it isn’t.
x0C, x0F, and xMF are quite interesting. x0C happens most and is sort of OK to buy, but x0F and xMF not so much.
The aggregated matrix is how I built the rules for the Specter/FF5 strategies. It is hard to visualize how to do that with consecutive time. Basically they buy the states that make money consistently.
I have the same result as yours by checking your signal for the past few weeks except today which shows as xFC.
Thanks for mentioning that, it looks like my problem. For intraday updates, I add the date to the SPY Signals spreadsheet and just update that new line with today’s prices. Yesterday I ran it once without doing that, thought I caught it before the update. I’ll start saving the intraday sheet to compare it to the official one after the close.
Date AVG PM13 PEM13
9/1/2020 351.5 0.0245 0.0218
8/31/2020 349.9 0.0231 0.0209
8/28/2020 349.8 0.0259 0.0242
8/27/2020 348.3 0.025 0.0239
8/26/2020 346.5 0.0232 0.0231
8/25/2020 343.5 0.0175 0.0184
calculation seems to be the same till 8/31. on 9/1 PM13 and PEM13 are higher than 8/31 and therefore, it is xF.
Yes, I agree with your numbers although we have different formatting ideas. Thanks again for bringing this up. I’m impressed with your work.
I’ve it build it Googlesheet. Thank you for your FSM idea. I’ve been thinking how to improve it but haven’t found any. One particular area I am looking at is the TrendFollow because It is not intuitive as compared to FSM.
Thanks.
There were a lot of personal achievements in the project. Basically the simplest explanation for the power comes from integrating above and below with up and down. That might be a real breakthrough in histogram analysis.
Trend following deserves more research. The theory is correct I think, but the methodology is more of a first attempt.