Tuesday, October 7, 2008

Revelations



As mentioned via a comment, the content of this blog has drifted away from the mathematical side of things for a good while now; I wanted to express some personal thoughts toward the financial industry at this point. Nothing is as it seems.

Pure Quantitative Studies are Over-Rated

The search for an elegant mathematical system to both explain the past and exploit future moves of the financial markets continues, yet so far I have not found anything of significant value to study historical prices. Too many assumptions regarding distributions, probabilities, rational trading behavior, and etc. have done more harm than good (e.g. consider the credit crisis and the half quadrillion derivative shenanigans).

So for anyone starting out and trying to figure out patterns, wedges, flags, trend-lines, let me save you some time and suggest that it is all a waste of time. Studying prices and price derived “dependent” variables will not result in consistently profitable trading.

Some Things that have “worked”

The news. Trading against mainstream financial media has worked exceptionally well. All you have to do is put yourself in their shoes, and think critically of motivations behind each public announcement, and more often than not obvious answers lie behind the veils of deception. Remember my article on Goldman Sachs and oil?

Quantified sentiment. Only by observing what most traders are actually doing, can one truly assess realistic likelihood of liquidity in the markets and therefore directional bias. One example of such would be the ISE sentiment index.

Credit risk assessment. An easy way to find good candidates for shorts is to look for a company that issues bonds of very high Yield to Maturity. The higher the YTM, the more willing investors “imply” high credit risk of the issuing business. At the same time, if the company is willing to pay such high interest for borrowing, it MUST be desperate for cash. This can not be faked by playful accounting.

Market depth of ECNs. Observing bid/ask volumes off BatsTrading or ARCA do provide a significant edge in day trading, unfortunately my current schedule does not allow for late nights every week day.

Existing research

My research this moment lies applying Support Vector Machines in neural networks for financial forecasting. Sounds impressive, though the mathematics gets so convoluted I get easily lost within the numbers, let alone trying to predict the markets for the next day or two. Nevertheless, I still search for something that “works” on the very short horizon.

I believe that method(s) MUST exist to determine high probability direction of stock indexes at any point in time, and finding answers from pure mathematics is improbable. From this point my interest has turned toward that of economics, politics, and behavior of the financial institutions, and this explains the change in the title.

4 Reflections:

alexandroid said...

I am glad to be your reader. =) Hope I am not the only one. =)

What inputs are you using for NN? Is it pure OHLC prices + volume or smth else? And what time scale you are working in?

I tried reading "Intelligent Systems and Financial Forecasting" but did not finish it yet since I've met notions of the facts that the most simple models work the best. Still, I did not find my good enough model yet. =)

Rocko Chen said...

Thank you very much for reading, that makes my day.

As for the NN, I try to keep all inputs as independent from the actual predicted variables (indexes). These include sentiment values (ISE, Investors' Intelligence Reports), implied volatilities (VIX, VXO), liquidity (CoT), and economic indicators like rate of inflation. Trying to pull it off in Matlab, but it hasn't been all that "successful".

Cutting to the chase, I find it easier and more "reliable" to just analyze the above variables manually and make decisions with my own brain. haha.

Rocko Chen said...

By the way, www.marketvector.com posts free 4-day forecasts off NN based analysis.

www.forecasts.org also do the same for slightly longer time frame.

Just places to reference, and I'm still a bit skeptical due to their pure quantitative nature, and they most likely used only historical prices as input.

alexandroid said...

Interesting link, thanks!

Yes, their current S&P 500 forecast was way off in Sep and likely to be in Oct. I guess that is a nice example of such models do not work good when markets are in transition...