Hello Mark, On Curve fitting.
My view on curve fitting may invalidate it's use for investment purposes as you would invoke a process that invalidates the underlying assumptions for curve fitting. On the other hand, if a certain technique [gives desirable results then I will be the last person on earth to object to its use. With curve fitting we attempt to model the real operation point of a process more closely then the measurements can reveal(due to errors). So. if data points are taken closely together then we can expect that the fitted curve matches the real data more closely than the measurements do.
In investment methods we already deal with the real data (for as far as stock prices are concerned) and in any attempt to fit a continuous curve trough the data points the real data is destroyed(as far as the fitted curve is concerned). This is the essence of my questioning the use of curve fitting with stock prices(Do a curve fit on a saw tooth wave and my point will be clear to all....you get the profile of a very worn out saw).
Mind you, I accept that important information for investing could well be a result of curve fitting, by extending the purpose of curve fitting beyond my current understanding of its purpose.
You also wrote:
This is why you should not use the AI 2.0 optimizer to optimize the parameters (i.e. create a Model) for, say, 1 year's worth of data and then use this Model for your 5 year investments. Rather you need to create a Model based on a number of time periods using different types of market conditions.
If your Model works well over all of these data points (i.e. prices), then you can be reasonably satisfied that your Model will do a good job in the future.
I read in this more a statement on optimisation rather that on curve fitting as such. If the optimisation works well then it is not relevant if curve fitting is used or not. So, I also accept that you might use optimisation to advantage. If the shoe fits, wear it!
What I object to is the argument that using 5 years of stock data works better for future system performance rather than 1 year worth of stock data. The rationale you use here, I think, is not so much an argument that a 5-year data span for an optimisation is more effective for predicting future performance than a 1-year data span, but that for may stocks a 5-year stock history usually defines the current stock behaviour better than only the stock history of the past year. Your statement might be based on your experience that for many companies the 5-year track record is more revealing for the future than the last year, but it certainly would not be true for companies that underwent a drastic change 1 year ago and have since exhibited a steady state performance on a completely different stock price. In that case the 5-year fit would be totally unfit for use.
What I conclude from this is that an optimisation process, (with or without curve fitting) must be selected based on what you already know about the stock history rather than systematically optimising any program on a 5-year-stock-data-run. Two examples Of what I mean are our KPN and KPNQuest, for which stock prises plummeted in the last few years(what else did not?). KPN will probably survive(price about 6 Euro when last saw it, and formerly abut 27 Euro). For KPNQuest, I just heard in the background, the latest deadline for survival has passed. Optimising these two stock for a 5-year history and with current low value is simply a mathematical exercise, but it would, of course, be futile for investment purposes.
I believe that in the discussions on optimisation(etc) the need for this type of thinking about an issue is maybe avoided too often, and perhaps more than is good for the discussion does the superficial views gain the greatest attention.
I hope that my treatment of the issues in this way is appreciated.
Regards,
Conrad
Market Data 
Markets 
AI
