I create models to help me estimate the odds of the market moving up or down over some period of time. Such models can be useful, but it’s essential to remember their limitations.
By their nature, such models only analyze past market behavior, and make their estimates by assuming future behavior will be similar to past behavior. However, past performance may not be indicative of future performance. I have tested many models that worked well for a period of time, then at some point stopped working altogether.
Models only give a statistical estimate of possible outcomes. Even when the model is providing good estimates, individual outputs can literally be anything. E.g., the model may be correct X% of the time in a given year when it predicts the market direction over the next 4 weeks, but still have a long series of such predictions where the market goes in the opposite direction. That’s the nature of averages. Averages and distributions don’t tell you anything about the outcomes of individual experiments. Incorrect model outputs are not necessarily evenly distributed, they can be and often are clumped together providing a long series of incorrect estimates.