Every so often when I was working in a bank, someone would ring up, and ask me what our 5 year EUR/USD forecast was. Trying to create such a long term “forecast” is indeed very difficult. One way is to use a valuation model such as PPP, which attempts to gauge long term fair value. However, even then, such an estimate might give you an anchor of where the price should be, but very little indication of the timing of such a move. In practice currencies can remain heavily overvalued or undervalued for long stretches of time, before a big enough enough event triggers the dislocation of markets and a realignment towards fair value. There is also the question of what fair value is, given that different valuation models are likely to give you quite different estimates.
More broadly we might want to ask the general question which types of forecasts are most useful for traders? We have a finite amount of time which we can spend investigating the market. Hence, it becomes crucial we spend time answering the right questions (sorry for that cliche) and focus our energy in forecasts which can be interpreted to make trading decisions. We can illustrate this with a simple example.
If we trading at relatively short time horizons it is basically irrelevant where a currency will be in a few years. Admittedly, this is an obvious point! However, what about very short term forecasts? We need to make a distinction between what type of thing we are forecasting and actual P&L we can generate. Ultimately a trader gets paid for generating P&L,and not for getting an accurate forecast. Let’s say that we can forecast nonfarm payrolls well. Our forecasts have the best standard error of any other forecaster out there on the Bloomberg survey. For sake of argument let’s say our standard error is 1k less than all the other forecasters out there. Is this valuable as a trader? Not necessarily. For a trader, being able to forecast whether NFP will be +225k or +226k is kind of irrelevant. Typically, if we are trying to judge economist forecasts, we would use standard error as a way of comparing them (and I suspect this is the way economists are ranked). However, for a trader is this the best metric? If we are trading on an intraday basis, what is far more important is to get the rough area where the number will be versus expectations. After all, if you are trading payrolls, what is important is whether you should buy or sell USD for your P&L, rather than minimising the standard error for our forecast.
Spending a lot of time and effort to get a forecast more accurate, when it won’t affect a trading decision is not going to help your P&L. It is much better to focus on forecasting which can materially impact our trading process. It is not always the case that a market model or more qualitative research will directly give you a buy or sell signal. However, we should before spending any time on research, try to think how we would use it when we trade. The same is also true when we look at new and exciting datasets. We might find them very exciting, but if we cannot coerce the dataset into forecasting something useful for trading purposes, maybe we should think twice about its relevance?