In this article, we give you a general description of the theory and technology of a Prediction Labs Inc. model. What are the fundamentals of this revolutionary new study? What can a trader expect by following this indicator? How does the system react to news? Here, we'll discuss these questions and others.
Predicting is about telling how things will look in the future within a certain margin of error. The margin must be small enough to make the prediction meaningful. Forecasting asset prices is a challenge that has fascinated investors since the very advent of financial markets; accurate predictions for the movement of financial markets can lead to fast and substantial trading profits.
There have been numerous attempts to forecast stock prices. Investors have used regression charts, fundamental analysis and indicators in an attempt to predict future stock prices. Most of these approaches are often not repeatable and lack a systematic way for measuring its accuracy.
Other approaches, like ours, model the financial markets from first principles.
We built a simplified market mechanism that assumes two different sources of power that move the market:
- News: We assume that this is the only source for a change in prices.
- Prices: We assume that prices drive themselves.
If prices are too high, sellers will drive the price back down. If prices are too low, buyers will push prices back up, even in the absence of news.
Market oscillations result from a detailed balance of expectations, generated both by news arrival and price changes.
Those who buy think prices will rise further, and those who sell believe prices will continue to fall. When the equilibrium breaks between those forces, a trend shows up. More subtle mechanisms are also at work.
For example, knowing that persistent movements in financial markets do exist, some investors act in consequence, causing these movements to be exaggerated.
This Prediction Labs Inc. model comprises a series of interaction rules among agents that make up the basic unit necessary to create a mechanism that reflects stock market dynamics.
To find the relationship between the price of an asset at any moment and the subsequent price, we can try to measure the balance of expectations. But, we cannot base the change in expectations on the news because news arrives randomly.
At Prediction Labs, we verified that the role of news in market behavior is often overestimated, mostly because investors need an explanation for a loss or a gain. The arrival of news is sudden and usually drives the market almost instantaneously. This is the starting point of a longer, autonomous, complex evolution where rules are written in the same prices.
Furthermore, we suppose that certain rules govern the way in which the price of a certain asset could evolve, plus random shocks from outside influences, such as news and events that we can’t foresee. However, if the shocks are not continually disturbing the market, we can glimpse into the future.
With the arrival of greater computer power, new methods are available to understand the intrinsic dynamics of the financial markets. Tools, such as chaos theory, the study of complex systems and dimensional shrinkage, are new approaches to studying the markets.
Many diverse systems in nature exhibit very complex, apparently random behaviors that can be appropriately described by simple equations.
Consider the human heart: It is a swarm of individual cells -- very similar to each other, although not identical -- interacting all the time. However, if you observe the heart as a whole, it possesses a harmonic behavior that can be described by few equations.
This drastic reduction in the number of variables, needed to describe a phenomenon is called "dimensionality reduction". These kinds of emergent cooperative behaviors are typical in systems driven by the aggregate of a lot of interacting individuals.
If we make the assumption that the financial markets are complex, auto-organized systems composed of similar individuals, all trying to maximize their income, we can hope that a description of low dimensionality may be suitable and that a certain forecasting capacity is possible.
The results of our research have made it possible for us to achieve a system that predicts, within a certain margin of error, the market path.
For example, in this chart, the market prediction is the black line while the actual market is the red line.
Example 1. SP500 July 13, 2004 (13:10ET).
This prediction was computed at 9:45 ET. At that time, our system charted the black line.
Example 2. NASDAQ100 June 14, 2006
(17:16ET).
This prediction was computed at 10:15 ET. At that time, our system charted the black line.
This displays the market on top of the prediction. We assume that stock prices reflect all the available information, so a price series should be enough to study the market. If we accept that the news arrives in a random fashion, we cannot waste any effort trying to predict these events. We must restrict our system and consider that it is randomly shocked externally.
Starting from the time series of prices and following a method proposed by F. Takens, we can build up a multidimensional trajectory. Then, we suppose that the observed evolution is determined by a system driven by news events and the free evolution of the intrinsic dynamics of the markets. Therefore, the actual movement is:
News-> Free evolution-> News-> Free evolution-> News-> Free evolution->
Our task is to try to understand the free evolution rules. Unfortunately, we cannot completely separate the data corresponding to free evolution of the market from the movement that is news driven. We must consider that the time series is polluted with external noise that we must try to clean.
A general question a trader always asks is this: How is the system performing when the market is affected by news?
As we all know, some news is clearly predictable, for example, an FOMC announcement.
The following is an example of how the market evolved with respect to the predicted path, on the day of an announcement.
Example 3. DJIA May 04, 2004 (18:00ET).
On May 4, 2004, The Fed left rates unchanged while indicating that the removal of its accommodative policy would be measured. In this case, the predicted path and direction of the market was correct from approximately 9:45 a.m. until 2:00 p.m. The impact of the news moved the market out of the predicted path. The market “reacted” to the news, but, then, one hour later, it recovered to the original path and closed at the very same price that was predicted at 9:45 a.m.
Our method of achieving this goal is to ignore the details of the movement and concentrate just on the major turning points of the price evolution in different time frames. Then, we numerically look for a set of equations that can account for this behavior, and we extrapolate this behavior into the near future.
Doing so, we can achieve accurate forecasts of the major market turning points without being side-tracked by external events.
Everyday, our system is able to identify pockets or predictability and trade them successfully.
Example 4. NASDAQ Apr 25, 2006 (16:07ET).
Statistical Significance of the System
These examples might look convincing, but, in general, the next two questions a trader may ask are these: How statistically significant is this system, and how much money can be made with these predictions?
To answer the first question, we use a statistical significance analysis and a back testing track record of a trading system working on top of the market's predicted path.
The outcome of the statistical significance analysis, which exceeds the scope of this article, is that the system does predict and has a measured edge because the relation between the Root Mean Square Error (RMSE) of the prediction versus the Root Mean Square Error of a random market is 71.2 percent. This means that, on average, the system gives the trader a mathematical proof edge of 21.2 percent over the 50 percent that one can get with chance.
This can be shown by the following chart.
In this chart, the red line is the RMSE distribution of the prediction while the blue line is the RMSE distribution of a random prediction over two years of data.
Trading System Back Testing
To answer the second question, back testing and forward testing was done, using the DJIA, the NASDAQ, NASDAQ100 and SP500 predictions.
The results obtained for a 30-month period are very encouraging. If you are a trader who follows the trading system signals without any other indicator, you can obtain an average of 47 percent annually to 350 percent annually, depending on the leverage you feel comfortable with.
Still, some traders might improve even these numbers by adding their own studies to our signal.
For more information on Prediction Labs Inc. technology, please visit our website at www.predictionlabs.com. |
