Saturday, March 7, 2009

That Sinking Feeling...

MATRADE released the January trade numbers yesterday, you can read the reaction in the Star here. I've added seasonal adjustment to the export and import numbers, but they don't look any better:

Since the y-o-y numbers are insane (Jan 09 exports are down 32% in log terms) and is probably affected by being based on export levels from last year, here's the month-on-month based on the seasonally adjusted series (log change from previous month i.e. non-annualised):

This looks just as bad.

At this point I'd like to discuss forecasting - given that exports and imports appear to move together, can one be used to forecast the other? Common sense and theory suggest you can - the structure of Malaysian imports supports that notion, where almost 70% of imports are intermediate goods used as inputs for export goods. You can interpret this two ways:

1. Higher demand for exports raises the demand for intermediate goods, which then also raises imports.
2. Turning the story around, higher imports indicate exporters are planning higher output, which means higher exports in the future.

So there's a good case for using (lagged) imports to predict exports. To start with, I regressed current exports against the previous month imports (sample range is Jan 2000-Dec 2007), both series transformed by natural logs. This is an exercise can actually be done in Excel: use LN() function for transforming both series, then use the slope(y,x) and intercept(y,x) functions to generate the statistical relationship.

Here’s what you should get:
Ln(exports)= 0.79+0.94 Ln(imports(-1))

The coefficients are all significant, i.e. they test out as being greater than zero; R-squared which measures the goodness of fit is a high 0.85, and the F-stat shows that the equation as a whole is significant. Because of the log transformation, this is actually an elasticity calculation and is very easy to interpret – a 1% rise in imports is associated with a 0.94% rise in exports. Notice I don’t use the word “cause”, because although a statistical relationship has been established, causality has not. So we now have one way to forecast future values of exports – except this regression is wrong.

Why that is I'll have to leave for another day, but I'll close this post with the predictions of this model and we'll see next month if it's correct. Based on the above equation, we have a point forecast of RM35,968 million in exports (non-seasonally adjusted) which represents a 6.3% drop from January, while the 95% confidence range forecast is RM43million-RM30million.

No comments:

Post a Comment