I have tweaked the model once more. As I gain more knowledge the model gets more sophisticated. Note this does not mean I curve fit. There is zero back testing in the model to calibrate parameters. The model changes as I get more sophisticated, and then I simply modify the equation. What I have noticed since the new update is, the model and SPX agree within two handles of each other. Now, one would think this is great, he has found the holy grail, right? Well, I ask you, even though I think I have a decent (or even nearly perfect) understanding of how the SPX is priced, what good does it do to know that the SPX is perfectly priced to within two handles? As far as I can tell, two handle difference could easily be attributed to noise and it is probably close to impossible for me to take advantage of it. So this creates the next conundrum. If NFV always says that SPX is essentially priced "correctly" to within two handles, what good is it? Allow me to answer that question with an analogy. My current thinking is that the market is like an organism, with genes that are constantly expressing or repressed at any given time. So say my model has three parameters, price of gold, price of oil, price of GE, and by some machinery, I take those three values and out comes a number. If I used the three parameters like this, SPX = GOLD + OIL + GE, implicit in that equation is that there are weights to each of those parameters, or more explicitly, a*GOLD + b*OIL + c*GE = SPX, in this case a = b = c = 1. So abc is the genome that gives how expressive each parameter is, in this case the genome is 111. If the genome was 101, the equation would become, SPX = 1* GOLD + 0 * OIL + 1*GE, or if the genome is .5.5.8, the equation would become SPX = .5*GOLD + .5*OIL + .8*GE. Now here is the point. My genes comprising the genome are all 1 at this point. The reason is that so far, having the genes be exactly 1 gives me the price of SPX to within two handles. I am actually hoping that this is false (not too much to ask for since I am 98% sure that as markets change the gene expression will need to change), and then the question becomes one of predicting what the gene expression will be "better" than the market currently expresses (this would then pass subtlety from biology to information theory). The point is that if my equation is right and there is no prediction (gene expression is always 1 and the list of parameters never change), then it is all hopeless, because that would mean that while I found the holy grail, it would mean that I would have no edge in trading it since the market would be nearly 100% efficient from the point of view of my model. Another possibility is that I am simply too dumb to understand how to make use of what I have now without having to wait for the gene expression that makes up my parameters to change. "NFV" ~1179.69 At the time of that I wrote that, SPX = 1,179.48. It has been tracking it almost exactly for three days counting, in realtime.
i couldnt program my cell phone so discard this if it's over simplified,but what if you added the transports into your value model,considering the dow theory
trying to understand: let's say a=b=c=1 for the last 3 days and NFV perfectly fits to SPX. if on day 4 NFV still tracks SPX perfectly but is off by -7 most of the day you will adjust c to maybe 0.8 to match NFV to SPX. but if on day 4 NFV diverges/converges with SPX, you will leave a=b=c=1. is this about right?
I would love to see it go -7 and then have SPX snap back to NFV. That would be edge, right? If instead it went -7 and then NFV snapped to SPX, that would mean that I am probably missing a parameter, or as you point out, since I am not varying the genes, that something is being expressed less or more than I am weighting. One other possibility on divergence that is only available to me is (since I am the only one that knows what the inputs and weights are) is to trade the spread between my model parameters and ES. First we need to see that I track it pretty well, and that when it does diverge, that they converge again within some reasonable amount of time.
Sure, that is a perfectly reasonable parameter. Remember though, I could get the SPX simply by taking every single component and adding the prices as per the equation that gives you SPX. The idea is to get some level of compression of explanation. The fewer the number of parameters that explain a variable, the more likely it is to be right. Occam's Razor type of reasoning (which btw, has foundations in information theory)
on every pullback in the spu's this last 12 months, the trans topped or weakened 1st,just a guess but if your model stopped running up before the spu's did ,would be a sell signal
the difficulty in backtesting NFV seems to come from the fact that several of a,b,c,d,etc need to have periodical parameter adjustment. i assume it is a process that is hard to automate. but what about doing a stress test: only testing NFV on ~20-30 days when some or all of a,b,c,etc make excessive moves. my point is that the way NFV behaves right now may not be indicative about its behavior under some rough conditions and you may not want to forward test it for months just to learn that it may totally break down one day.