Quality Calculation

The quality of the nets on your PC and the quality of the forecasts on the Internet pages are two completely different matters. The quality of the nets (visible in the Upper Left-hand Screensaver Honeycomb) is calculated on the basis of past control data and shows on a defined scale how well the net performs. In contrast, the forecast probabilities in percentages on the website show how probable it is that a particular forecast applies within limits of a particular tolerance level. It is calculated from the net's future (i.e. based on actual forecasts).

About the Net Quality Calculation (Upper Left-hand Screensaver Honeycomb):

During the net's training period, the quality curve rises. The forecast quality is measured on the basis of time periods randomly selected out of all the stock data. The net is applied to these data and the variance from the actual historical value is calculated in %. Only mistakes smaller than 10% are displayed. The curve is squared, since movements can be detected more easily at a higher quality. For the mathematically interested, the precise formula reads exactly as follows:

Whereby x equals the average variance between the forecast and the actual stock value %, f(x) then equals the quality of the calculation.

The Evaluation of the Forecasts on the Website in %:

The forecasts themselves are simply given in percent. The hit ratio of the forecasts (quality), a measure of reliability, is also expressed in two percentages, directly in the column behind the respective forecasts: two two-digit numbers xx/yy, whereby xx represents the tendency hit-ratio, and yy the corridor hit-ratio in percentages. Caution should be observed, however, when using these values. They merely illustrate how accurate our forecasts were in the past - if, for example, fundamental changes have taken place in the meantime, the values will look different in the future. Yet, this is a good indication of how reliable the quality is which can be expected in future. As or the individual values:

The tendency quality (front value) is fairly simple. It tells us how many of the forecasts of the past 25 trading days in the simple categories "rising" and "falling" were correct. We implemented the tendency quality because for most users, it is a more tangible value than corridor quality (see below). Were we to throw dies or simply guess, the tendency would always appear at a value of approximately 50%. All larger values count to the forecasts successes of the nets. We have be very strict: if a forecast predicts a rise of 0,1% where we predicted -0,1%, we have to evaluate the tendency as incorrect - despite the fact that the net is reliable, since normally a variance of 0,2% is quite insignificant. But we have to draw the line somewhere. This, by the way, is the rare case of the tendency quality pointing to "no" when the corridor quality control points to "yes" - the so-called sideways corridor (-0,3% to +0,3%).

The corridor quality (number behind), in contrast, shows how accurate the forecasts are. In this case, the net (as in the upper left-hand honeycomb of the screensaver) is no longer evaluated, the forecast is - the standards are more strict here. What is evaluated is how many of the forecasts (in %) were applicable in the previous 25 days. First, however, what needs to be defined is what can be accepted as "applicable", i.e. tolerance.

Forecasts: The corridors

We have structured the forecasts according to different categories: besides the sideways movement (i.e. slight price movement), there is a slight, medium, and a strong increase or decrease, in other words seven levels. A forecast is considered applicable if it hits the correct level, whereby hits in the neighbouring corridor count as semi-hits. The corridors are defined according to actual market movements - they were defined on the basis of our historical data in such a manner that each of the seven corridors contains on average one seventh of the hits. In this sense, the evaluation "slight, medium, strong increase or decrease" or "sideways" is as objective as possible. A slight movement of these probabilities obviously results from the differing volatilities of the values (e.g. more extreme hits in the Nemax), but it would be far too complicated to adapt this to each case - one would run risk of losing sight, and would need to adapt the values constantly (which lead to a loss in comparability!) and also have to provide page after page of explanation. That would not be feasible.

The probability of hitting the right corridor is 1/7. If the neighbouring corridors are included in the calculation, then a statistical hit probability of 13/49 results for the corridor quality, since the extreme values (strong increase or decrease) only have one inner, neighbouring corridor. On the whole, this approximates a statistical corridor quality of 26,53%. All higher hit percentages are considered forecast successes of the nets.

The Colour Codes: Explanation section contains details on the colour coding in the tables that follow the corridors.