Forecasting or nowcasting?

We all want to forecast what’s around the corner? However, perhaps a more pertinent question, which we might seek to ask is the following, is this actually a corner? In a nutshell, that is the difference between forecasting and nowcasting. Forecasting is all very well and good, but it isn’t particularly easy. Nowcasting is more about understanding the current situation. It’s all very well wanting to forecast GDP in a year’s time, or where EURUSD might be, but ultimately, this should be a secondary consideration to understanding the current state of the market. If we don’t have an effective model for understanding the current state, then it seems unlikely that we’ll be much good at forecasting into the future.

  

This has been of particular relevance over the past year or so, where traditional economic data has heavily lagged the gyrations of the economy, which have been impacted significantly by factors such as lockdowns. The solution is nowcasting, trying to get a handle on economic variables, such as GDP on a high frequency basis. Marcos Lopez de Prado and Alexander Lipton wrote about this in March 2020, in their note Three Quant Lessons from COVID-19. Alexander Denev and I also wrote about nowcasting in The Book of Alternative Data.

 

Working with economic data is often quite challenging, as I wrote recently, and we have to be careful when looking at timestamping, making a distinction between the date of release and the period which an economic variable covers. There’s also the additional point that there can often be many different revisions of the same economic data point.

  

One of the key components of a nowcast is to use alternative data. That isn’t to say we totally remove traditional data from the picture, those datasets are still important. It is more about augmenting our process to include alternative data, in particular many of these variables may give us additional information at a higher frequency, compared to traditional datasets. These alternative datasets might range from consumer transaction data, which is available with a relatively short delay, to mobility datasets, like those from Google and Apple, which have quickly picked up the impact of lockdowns. However, there are many more which funds are likely to be looking at.

  

We’d all love to forecast very far ahead. However, it is important we begin with understanding the current situation, nowcasting is an important exercise.