Weather forecasting will show the most likely weather for the coming five days, with a probability of its occurrence. For example, a 60% chance of rain means that, under the current conditions, 60% of the time rain will develop and 40% of the time it will not. If you look back at forecasts the predict a 60% chance of rain, it has rained in 60% of the cases.
Would it be possible to create a model that will forecast stock price movements over the coming month? It would be a probabilistic model, similar to weather, meaning that it will be wrong a portion of the time. But the idea is to quantify what that proportion is.
The most basic model predicts the outcome using no model at all, only historical outcomes. As an example, it would be like looking at the Calgary climate, finding that there is rain or snow on 152 of 365 days, then producing a forecast for a 42% chance of rain (or snow). This forecast could be considered accurate, but it’s not very helpful. A similar forecast for stock prices over the coming month, based on monthly returns since 1979, is a 59% chance of positive returns. In 59 months out of 100, stocks have experienced a positive price change. Given this fact, owning stocks is profitable (as opposed to gambling at a casino).
Next, I’d like to refine my model. I haven’t gotten very far. I have found that the current return to bonds is not predictive of future stock returns. The current return to stocks is slightly better. If stocks went up over the past month, there is a 64% probability that they’ll go up over the next month. It is possible to improve returns (ignoring fees and taxes) by only owning stocks in months when the previous month return was positive. I’m not suggesting this is a useful strategy, only that it supports the improvement to the forecast.
There are a number of other variables I will look at. But the noise-to-signal ratio is so high that it’s very difficult to predict future stock prices based on any single or group of inputs. In that way, it’s like trying to predict the weather more than a week in the future.