Sunday, February 8, 2009

To Be Or Not To Be: GDP, Seasonal Adjustment and Calculating Growth

I came across this article on China's 4Q GDP growth in The Star the other day, which claims:

"The government says the economy grew by 6.8 percent in the final quarter of 2008, but that is based on an outdated system that measures growth against the same period a year earlier...Compared to the previous quarter, the method used by most major economies, growth was as low as 1 percent and possibly zero, economists say."

Beyond the debates regarding how China is actually doing, this illustrates two different methodologies for looking at GDP growth. From my point of view they're both relevant, and in fact when you look at actually statistical releases, both data points are generally published. I would NOT however call the quarter on last year's quarter methodology as "outdated". There are solid grounds for using this methodology as the main point of reference, just as there are weaknesses. As pointed out, it doesn't capture the momentum of the economy very well. On the other hand it's better at abstracting from "seasonal" effects than a mechanical seasonal adjustment of the raw numbers would. How so?

The problem with applying seasonal adjustment to China data, as for any majority Chinese country, is the enormous impact of Chinese New Year on spending and economic activity. Since CNY follows a lunar calender, the effect falls earlier every year. For Malaysia, we also have the additional effects of Aidil Fitri, which also depends on a lunar calender. This presents a problem for seasonal adjustment since the various methodologies use historical data to calculate the seasonal indices. Just as important, adjusting for strong seasonal effects introduce an element of artificiality into national accounts interpretation. In practice however, these factors don't seem to matter much, at least in the data I've been looking at. What does matter is the q-o-q methodology tends to accentuate momentum too much.

To illustrate this, let's take a look at Malaysian RGDP (2001:1 to 2008:3; 2005 prices):
From the chart, the seasonal effect is obvious and persistent. This is what it looks like after seasonal adjustment (using the US Census Bureau X11 multiplicative method):The different seasonal adjustment methodologies yield broadly the same results, so I'm sticking with X11 for now - so far so good. This is what the different growth methodologies actually show for Malaysian RGDP growth:
As you can see, the growth estimates under the q-o-q method is much more volatile than the y-o-y method. The drop is sharpest between 2002:3 to 2002:4, from 12.0% to 2.7%, while other sharp swings are pretty clearly shown especially in the early part of the series. Even using seasonally adjusted data y-o-y shows the same results as the non-seasonally adjusted data, so it's more the growth calculation than the seasonal adjustment. Given this volatility, I suspect that for developing countries at least, the q-o-q method would be considerably less useful as a guide to policy than y-o-y. Even then, take a look at the same comaprison for the US:
The volatility is still there. I'm not convinced this is a better method of showing the performance of the economy, especially as a reference for monetary and fiscal policy. It's too easy to overshoot policy adjustments.

No comments:

Post a Comment