Tuesday, November 3, 2009

What The KLCI Says About Economic Policy

Usually it doesn't.

Teoh Kok Lin makes that point today (emphasis mine):

"BUDGET 2010 was, shall we say, not that warmly received by the man in the street. Some pointed out that the budget gave few goodies to the rakyat or businesses compared with previous years, hence it was not as good and that’s why the stock market was down 0.5% last Monday.

I believe a slight disappointment with the budget played a small role, if any, in the stock market. However, I am perplexed how one can infer from “one-day” stock market movements whether the budget is good or not."

To be more rigorous about it, here are the stats on the log difference in daily closing for the KL Composite Index (sample: January 2000-September 2009):

The two stats of interest are the mean (0.0001) and the standard deviation (0.0096). The observations show that daily changes are approximately normally distributed somewhat more leptokurtic than a normal distribution (from the Jarque-Bera stat "peakier" with a kurtosis stat > 3), with a mean of zero:

Using a 95% confidence interval and the standard deviation of 0.0096, we get:

(1.96 x 0.0096) x 100 = 1.88%

Which means that any movement plus/minus 1.88% is just the normal daily trading range for the KLCI, and doesn't say anything at all about the budget, RPGT, credit card taxes, fuel prices, commodity prices, the US dollar, Tun Mahathir's comments, corruption, politics, or Samy Vellu's hair (or lack thereof).

Edited (changes in blue): The actual distribution of changes has a greater central tendency than a normal distribution. Thanks Hafiz!


  1. wouldn't it mean that we should reject that the variable is normally distributed since the p-value of JB-stat is 0 (assuming the critical value is 1%)?

    I think that -0.10 data point is causing you problem.

  2. anyway, yea, ppl tend to read too much from the stock market.

  3. Thanks for the catch, I've amended the post to reflect that. My bad for doing the post in a hurry and not double-checking.

    The JB stat shows non-normal because the distribution is "higher" than a normal distribution. If that outlier was important, than you'd get greater skewness - it's the kurtosis that seems to be the culprit here.