Monday, October 12, 2015

Nowcasting and trend-following - cut from the same cloth?



There has been a movement in forecasting to better exploit information in real time. To a degree this has always been done on Wall Street, but the process has become more formal with structure on how information can be employed to enhance short-term forecasts. This process has been given the name - nowcasting which is the contraction of "now" and "forecasting".  It has been effectively used in meteorology and macroeconomics especially with central banks who want to get better real time estimates of the key macro-variables.

We think this forecast movement is a major advancement in thinking that can be more effectively incorporated in global macro investing. There is no secret formula for nowcasting; rather, it is an approach on how to use all available data that may have different frequencies, look-back periods, and sampling error. It is the use of ensembles of forecasts with different time scales and information quality to generated the best short-term forecasts.

One of the key issues with macro forecasts is that many of the key variable used are faced with delays, are infrequent, and subject to sampling error. For example, GDP comes out once a quarter with a delay but a lot of the information that comprises GDP is available sooner albeit in different forms. These data can be used to better estimate GDP to give an immediate forecasts.

Hence, if you use macro variables to generate signals on what you should do, there will be little action that can be taken between fundamental data points. There are no short-term forecasts. If the data used comes out monthly, you can only make a forecast once a month, yet markets trade everyday.  Nevertheless, investors actually have more data available that can be employed to make short-term forecast. Some comes out weekly, other come out monthly but may have less delay in its look-back.

In some sense, trend-following is a nowcast using just price information. In fact, you can see why trend-following has appeal as a nowcast for trading. You can get a short-term forecast immediately through extrapolating recent price activity. The nowcast from a price based system will be to use trends with different look-back periods to provide different types of information, short, medium, and long-term. The can be weighted to form a better forecast.

You don't have to wait for data on unemployment or GDP if you nowcast using price. However, if you truly want to use all available information, price data may not be enough. A more sophisticated nowcast will be to use both price and fundamental information as an ensemble. In the case of stocks, the nowcast can look at price trend with earnings information. The bond trader will use price, yield curve, central bank, and economic fundamentals. The commodity trader may combine price with forward curves, and inventory information. Nowcasts can use different information blends to make better forecasts.

While we have just scratched the surface for what is a nowcast and how it can be used, the concept can better focus the thinking of any macro-trader.  The integration of different data with different frequency is a better use of computing power than perhaps looking for the new best model in one time frame.

No comments: