Thursday, February 18, 2010

Panels and Pools: Malaysia's Demographics Part IV Continued

[Continued from last post]

Model III

In the last two attempts to make use of empirical evidence to suss out the relationship between the median age and dependency ratios of the population and income, we’ve looked at essentially two dimensional representations – Malaysian data across time, and a large sampling of countries at one point in time.

This post looks at a three-dimensional approach, of a large sample of countries across time (sorry, no graphs here). This is generally known as a panel or pooled approach, so if you ever come across the term, you’ll know what it means. Two further terms you need to know are “balanced” and “unbalanced”, which refers to data that is symmetrical (all countries have data for the sample period in question) or not.

Bear in mind that this is my first stab at working with panel data, so if you have any advice/contribution/criticisms to make, sound off in the comments – I want to hear from you!

I’ll skip all the technical stuff and go straight to the results (read the accompanying notes if you’re interested in the details):

  1. LOG(GDP_MYS) = 0.87 + 3.90*LOG(AGE_MYS) - 4.04
  2. LOG(GDP_MYS) = 0.66 - 2.11*LOG(RATIO_T_MYS) + 7.47
  3. LOG(GDP_MYS) = 1.16 + 1.76*LOG(RATIO_O_MYS) + 12.66
  4. LOG(GDP_MYS) = 0.89 - 1.83*LOG(RATIO_Y_MYS) + 7.17
  5. LOG(GDP_MYS) = 0.87 + 3.93*LOG(AGE_MYS) + 0.02*LOG(RATIO_T_MYS) - 4.12
  6. LOG(GDP_MYS) = 1.00 - 0.96*LOG(RATIO_Y_MYS) + 0.49*LOG(RATIO_O_MYS) + 1.69*LOG(AGE_MYS) + 3.54

I’ve only reported here the results for Malaysia; all the coefficients and regressions as a whole are statistically significant. As you can see (if you haven't gotten cross-eyed!), we again have general confirmation of the stylized facts:

  1. A higher median age is associated with higher incomes
  2. A lower dependency ratio is associated with higher incomes
  3. A lower youth dependency ratio is associated with higher income
  4. A higher old age dependency ratio is associated with higher income

The regressions that combine age with dependency ratios are less clear (just like in my first two attempts), with the possible interpretation that age has a bigger impact than any of the ratios. Unfortunately like in Model II, I'm still not getting plausible estimates for income - or I've screwed up somewhere. If I can figure it out, I'll revisit this topic in a future post.

It turns out that only the first, simplistic model (Malaysian data across time) comes close to providing reasonable estimates of income levels into the future. This suggests that the country specific factors driving Malaysian income are strong enough to overcome some of the disadvantages we face because of our demographic structure.

I have to reiterate this - Malaysia is a middle income country with a population structure of a low income country. I suspect that being an oil producer is one big leg up over that particular hurdle, although I haven't been able to statistically prove this - using a dummy for oil production across the whole sample didn't yield statistically significant coefficients. Other potential boosters that have generally been put forward in the literature include the ability to produce sufficient food (arable land) and institutional factors that have been the basis for inward investment.

Whatever the causes may be, we're ahead of the game in some respects and have the potential to sustain above-average growth going well beyond that of our regional peers - Malaysia's labour force as a proportion of the population will keep on rising for decades to come.

Technical Notes
  1. GDP data from the IMF World Economic Outlook Database (April 2009)
  2. Population estimates from the US Census Bureau International Data Base
  3. Sample size is 179 countries, with age, ratio and gdp data from 1991 to 2008 (unweighted, unbalanced panel; 3097 observations)
  4. GDP data is in current international dollars
  5. Fixed effect dummies are used to model individual country variations (second intercept term in the equations above) - no time effects (fixed or random) are assumed


  1. hishamh...

    visually ur scatterplot in previous post shows near reverse pattern between young and old dependencies ratios impact to income....

    Got a strange feeling on ur old age dep vs income..

    just a visual hunch

  2. That's the basis of one of the findings in the IMF article I linked to in the first demographics post:

    There's actually two demographic dividends - the first stage when you have a bulge in the working population. The second comes when this bulge moves into retirement, as they have the accumulated savings and assets to continue consuming and investing.

    That means the relationship between income (which here actually means aggregate output) and the old age dependency ratio isn't straightforward.

    Whether the relationship is positive or negative (as it is with youths) depends on how much these retirees have saved and their propensity to consume, and whether the country in question actually benefited from the first demographic dividend.

    The IMF article argues it's positive.