On Forecasting

Author: Emre Deliveli  ·  November 13th, 2011  ·  930http%3A%2F%2Fwww.economonitor.com%2Femredeliveli%2F2011%2F11%2F13%2Fon-forecasting%2FOn+Forecasting2011-11-13+18%3A26%3A20Emre+Delivelihttp%3A%2F%2Fwww.economonitor.com%2Femredeliveli%2F%3Fp%3D930 › Share This Print

I recently got an email from a reader, asking me if I had forecasts for the following variables: GDP (USDbn), GDP growth (%), Consumer Prices (%). Exports (USDbn), Imports (USDbn), Trade Balance (USDbn), Current Account Balance (USDbn), Current Account Balance (% of GDP), FDI (USDbn), TRY/USD, CBRT Policy rate. And not only for 2011, but also for 2012 and 2013… The last two are quite entertaining: When asked about his EURUSD projections, former IMF chief economist Ken Rogoff is rumored to have said “HTF do I know?”, which is, IMHO, the most accurate forecast he could give. As for the policy rate, the Central Bank’s latest actions have made sure that it doesn’t matter anymore…

Anyway, I kindly told her that I am crazy, but not that crazy: Since I left being a market economist back in 2008, I have no incentive to produce detailed forecasts. But given that I am your friendly neighborhood economist, I felt compelled to provide some guideline to my reader to help her differentiate between all the forecasts she is gathering:

1. There is no perfect forecast. Sure, VAR may beat simple exponential smoothing or linear regression, ARDL VAR and DSGE ARDL, but as Koc University ERF Director Sumru Altug noted at a chat a couple of weeks ago, it would cost a couple of years of work  and more than a million bucks to get a good forecasting tool and even that will not yield perfect out-of-sample accuracy. So use whatever method you want to use and test the out-of-sample accuracy. I would also urge you not to get stuck with standard metrics such as MSE, MAE or MAPE: If you will use this for trading for example, getting the correct sign of USDTRY is more important than  a very small MSE that gets the direction wrong half of the time! All these points (and more) are in the lecture slides on forecasting of a time series course I taught a couple of years ago.
2. Although there is no perfect forecast, I still would want to know about what the forecaster has done. You almost never get methodology descriptions in research notes of market economists, but I always look favorably on the ones that at least have a brief description. I made that point over at the old blog quite some time ago. My friends at YapiKredi Research maintain that forecasts should be a statistical statement reflecting the methodology (growth will be between 4.6-5 percent with 90 percent probability) as well, but once you go down that route, I doubt anyone will make the cut:)
3. Call me mad cap, but unless someone has done something completely crazy, the internal consistency of the forecast set is much more important for me than the methodology. What does this mean? Someone with some basic macro background should not look at your numbers for ten seconds and say, “what the hell have you been smoking?: Again, there is a vivid illustration at the old blog.
4. So how will you ensure internal consistency? You just need to know a bit of the different relationships of the macro accounts. Then, you forecast a couple of key variables, with VAR, DSGE or throwing darts and build the other forecasts around those. If you have some macro background, just go through a simple exercise I used at a class once (don’t look at the solutions and explanations before you have worked on it for a couple of hours). And if you are really into this, I would recommend you a somewhat old, but still useful IMF Financial Programming book, which uses Turkish data nonetheless.
5. Finally, I would not make much of anyone’s 2013 forecast, especially this early.

Have fun!!!!:)…