Friday, December 14, 2012

The VIX

Pretty much 100% of my trading time is now spent on VIX futures.  There are philosophical reasons for this approach at this time which are too long to detail here. The summary is that the VIX seems to be relatively immune to the noise generated by HFT and stop-loss discovery algos. I abandoned structural modeling for this project and used a neural network, formally an "SVM" or support vector machine for data crunching.  I have been long since 2012.11.21 at the close  and it looks like today the output suggests continuing a long position.  For the next month, I will post daily updates here on what the VIX model indicates. This is of course purely for amusement purposes.

I have found the stop price to be reasonably well estimated by taking a moving standard deviation of the time series over 8 days, and subtracting three times its value from the previous closing price.  Wash sales are highly annoying!

2012.11.21 15.31 BUY at the close
2012.12.13 16.56         stop @ 15.39

2012.12.14 Monthly Gold Cycle Chart





















Here is the new gold cycle model chart. As with the silver model, all cycles < 233 days were excluded from the analysis, as well as any cycle information obtained from outside the data series.

2012.12.14 Monthly Silver Cycle Chart






















This is the new cycle model for silver. It has been pruned so that all data inputs are directly related to the time series and nothing else. All cycles with periods less than 233 days were also eliminated from the analysis because short-term fluctuations were much less helpful in decreasing error than has been the case in the past.  I'll see if I can update it at least monthly. 

2012.12.14 Monthly DJIA Cycle Charts


































I have been slowly rebuilding the cycle models, which are now less complex as a result of the input data pruning. As promised, here is the new cycle model for the DJIA.  One of the cycle model inputs, a cyclical model of its own relating the yield curve spread and rate of change of US monetary aggregates is worse than useless and was contributing to the increasing error. I have also dropped all inputs regarding employment data from the US BLS. The increasing second-degree fluctuation of their short-term variance convinced me that the set of data I was using contained more noise than information. The academic question is now not whether  Fed interventions can change business cycle amplitude and phase, but by how much and over what time window. Right now, it looks like the long-term cycles with a period > 8 years are stable since 1896, and it is the noise in the shorter cycles that created the error in the previous model version beginning around early 2010.   So according to the model, we might have topped out on the DJIA.