Annotated Models
This site is best viewed in a browser that conforms to web standards.
We are not presenting the strategy model for Berkshire Hathaway as a recommendation for managing the stock with NeuroStrategy, even though the strategy developed would invite it. We are showing this model primarily in order to demonstrate that NeuroStrategy's methods of learning and backtesting are independent of the actual price level of a stock. Technical trading supported by artificial intelligence works best with securities, for which sufficient liquidity exists, and where the unit price is within a "normal" range. NeuroStrategy's risk management reacts to uncertainties by adjusting allocation sizes. In order to be able to follow the strategic instructions, the initial investment should be chosen in such a way that in case of a recommendation of 100% long positions, one is able to purchase at least 100 units of the relevant instrument. This is the only way to ensure that on subsequent days, one is actually able to reposition as recommended (to 80% long, for example) by selling 20% of the stocks. In the case of stocks in the price ranges of Bershire Hathaway, this is a problem.
As the following chart shows, the high unit price does indeed affect the considerations regarding the initial investment. The actual price, however, has no impact on the learning behavior of NeuroStrategy. NeuroStrategy converts the price information into a form that makes the system independent of actual price levels. A simple example of such a conversion is the representation of the time series through day-to-day deviations expressed in percent. In this way, all information is preserved. Yet, it no longer matters whether the instrument examined is a "penny stock" or Berkshire Hathaway.
The model for NYSE:BRKa appears very robust. There is no detectable slump in performance during the period of productive use. This means that NeuroStrategy has succeeded in developing an intelligent strategy that performs just as well in the trading environment of the real world as it does in the protected learning environment, where the teacher knows the answer to every question. In actual production, of course, we do not know the answer to the question, "what should be done?". Rather, in this environment, we use the software to help us answer this question.
Even though the equity development in the above chart appears positive, a closer look at the period of production reveals features that urge caution. Although the strategy clearly outperforms the development of the stock, the model shows little flexibility in dealing with its mistakes and reacts rather hesitantly to price changes. The volatility almost corresponds to that of the instrument itself. That is, NeuroStrategy often takes full positions.
The signal chart confirms this observation. The system reserves its resources for correct predictions, on the basis of which it then decides in favor of a change of position. This is a completely different approach to what we find in the case of the strategy model for AOL, where risk management and adjustments of position dominate the picture. Thus, the chart for BRKa gives the impression that no asset allocation is involved. De facto, the difference in performance only amounts to 1%, if in this model one merely reacts to signal changes. This explains the higher volatility of the model: it prefers to wait and see rather than do a lot of trading. This approach also seems practical. After all, with the high price of the stock, a trading transaction at a loss immediately has very painful effects. The focus on correct predictions, however, proves to be just the right solution. Accordingly, the model is very accurate and is usually positioned correctly in terms of long or short. In case of a mistake, the stop-loss is set at a narrow tolerance and the opposite position is taken rather consistently.
This procedure has fatal consequences for investors who must take transaction fees into account. The following chart shows what the equity development looks like, if US$ 5.00 are incurred in fixed costs and 0.18% in broker's fees per transaction. Due to the high unit price of the stock, broker's fees play a much larger role than the fixed transaction costs. It also becomes clear why NeuroStrategy "decides" on the strategy displayed. A risk management as in the case of AOL with frequent adjustments of position would be disastrous in the case of BRKa. NeuroStrategy must rely on its predictions and - if it detects a mistake - it must change positions completely. This way, the transaction costs incurred are lower in absolute terms, but they are still high enough to eat up any possible profits gained through correct repositioning.
Thus, traders who have no or only fixed transaction fees to consider can use the strategy model without hesitation. Investors who must pay broker's fees, however, should stay away from this stock.