In meteorology, with nowcasting you want to perform short term estimates and prediction of thunderstorms or other events. Typical questions you want to answer are: where is the storm now? Where will the storm be in 60 minutes? What is its path? Is it weakening or strengthening? Are there hazards associated to it (lightnings, flooding, etc)? The problem you want to solve is about spatial location and intensity. It is similar to what we want to know about price: will it be up or down in X hours? How many points? Predictions are performed using algorithms and extrapolating radar echos (based on movement) assuming no growth/decay, which is normally acceptable for 60 minutes forecasts. There are several methods to track objects and estimate motion in real time using filters (e.g. Kalman) and calculate trends. The longer the time considered for the prediction, the higher the uncertainty. Accuracy is strongly depending on the type of model you use. In the stock market, when you try to predict the price of an asset at a certain time in the future you take into account the information you have at the moment you do the forecast. This involves uncertainty, which is also function of the time horizon of your forecast. As time progresses, you are able to include new data and more information in your model to update and refine your forecast. With time, you have more data available and your forecast horizon decreases. As uncertainty decreases, you have a more accurate forecast, but also have lower margin profits. For example, at the close today you forecast the next day's close taking into account the data available up to that moment. As time progresses, you include in your model the overnight action, then the open, the morning action and so forth getting closer to the end of the trading session. Accuracy will depend on how your model smooths data and extrapolates movement.

How can we use this methodology? When you open your trade you expect a certain behavior as time progresses. Nowcasting can be used to check if the path (price) of your asset follows the expected pattern. You have a number of "what if", such as: what happens if it opens higher, what happens if it prints a spike to the upside and so forth. Each time you update your expectations and when the expected path deviates too much from your original forecast it is time to close your trade. Conversely, if, with more data, expectations are reinforced you keep your trade going. It is a matter of thresholds of course and accuracy of your model, that depends much also on the frequency of observations and amount of data available.





Speak your mind


Resources & Links