Wednesday, July 10, 2019

More fun With the CPI


There are interesting things hidden in the depths of the monthly CPI report.  Remember that the CPI-Total inflation rate is calculated as a weighted average of the rates of inflation of individual commodities and that any time you’re working with a weighted average, the overall average will change if the distribution of weights across items changes, even if the values of the items being averaged don’t change.  In the case of the overall (CPI-Total) inflation rate, the weights are the shares of total consumer expenditure devoted to each commodity, as found in Statistics Canada’s household expenditure surveys, so rates of inflation for commodities which take up a larger percentage of the consumer’s expenditure get greater weight in calculating the CPI-Total inflation rate. 


Those shares are selected to be as representative as possible of the average Canadian household, but clearly there is a wide range of different types of household in any population.  How well the CPI-Total inflation rate reflects the cost of living experience of any sub-group in the population – young families, for example – depends on how closely that sub-group’s expenditure patterns (i.e. the shares of their consumer dollars which they allocate to different commodities) are represented by the weights that StatCan derives from the survey.

These days the household expenditure survey is pretty detailed, meaning that StatCan asks a lot of questions which let you look at the expenditure patterns of different types of households.  So you can start from the full survey data set, and pull out just households whose incomes fall into certain ranges.  Given that subset, you can calculate what proportion of their total expenditure those households spend on each of the categories in the CPI basket.  Then you can take the CPI data on the commodity-specific rates of inflation, weight them by the expenditure shares for the population subset which you’re interested in, and come up with an overall inflation rate which applies to the group you’re interested in, and which can be compared with the CPI-Total figure (remember that the commodity-specific inflation rates are the same in both calculations, it’s the weights on each which differ). 

Statistics Canada recently did that sort of exercise to come up with a CPI aimed specifically at calculating the rate of increase of the cost of living of Canada’s senior population. 
On a related issue, Herb Emery of the University of New Brunswick had a piece in the National Post recently in which he argued that the total CPI figure underestimates the increase in the cost of living for lower income Canadians, among other groups.  (The technical report is available here.)

Herb took a different approach to the question, although he also used data from the survey of household expenditure.  His starting point was Engel curves, which show how the share of consumer expenditure going to various commodities changes as income changes.  There are certain regularities which come through in the whole body of research on expenditure shares: basically, lower income individuals devote a larger proportion of their consumer expenditure to necessities than do higher income individuals.  If you like, think of that the other way around – higher income individuals devote a higher proportion of their consumer expenditures to luxuries than do lower income individuals.  That’s pretty much inevitable, since the essence of necessities is that you buy them first, and only buy luxuries if you’ve got enough income left over after buying the necessities.  That doesn’t mean that you spend fewer dollars on necessities as your income rises, but you spend more dollars on luxuries, and since shares have to add to 1, the share of expenditure which goes to necessities declines as income rises.  Some of those regularities are more regular than others, and the one which Herb used relates to the share of consumer expenditure devoted to food, one of the most consistent of the regularities.  Lower individuals devote a higher proportion of their expenditure to food than do higher income individuals.

Obviously a whole mess of factors come into the family’s decisions about expenditure on food: food prices, for example, and family composition.  But controlling (statistically) for those we find that the share of expenditure devoted to food declines as income rises.

Herb used this result to try and estimate how well the CPI-Total inflation rate, which is based on a statistically average family, reflects the cost of living of families in various sub-groups of the population. This is where the reasoning gets tricky.

Remember that the idea is to evaluate the CPI-Total as an indicator of the cost of living of (let’s say) lower income families.  We’re going to do this using data from several years’ worth of the household expenditure survey.  We start from the observation that in the first year of our data set, families in our income range allocate a certain proportion of their expenditure dollar to food.  Now suppose that over time, nominal income rises but all prices rise at the same rate, so that relative prices don’t change and the cost of living rises at the same rate as nominal income rose, meaning that real income (nominal income deflated by the CPI) hasn’t changed.  In this case, even though nominal income has gone up, the share of expenditure which goes to food won’t have changed.  All that will have happened is that the number of dollars which our target group spends on food will have increased in exactly the same proportion as income increased. 

Remember what we’ve got to work with – the share of food in consumer expenditure, which is a percentage; nominal income which is in dollars; the CPI-Total, which is an index number; and real income which is nominal income deflated by the CPI. 

Now suppose we plot our data and find that real income has remained unchanged from year to year even as nominal income has risen.  That means that the CPI-Total has risen at the same rate as nominal income.  But suppose we also plot the food expenditure share and find that it has changed, when we had been arguing that it should remain constant.  And suppose that we’ve controlled very carefully for changes in things like family composition and interest rates and anything else which might have affected consumer spending decisions.  Something has clearly gone wrong, the question is, what?

The key to understanding what’s going on is the fact that we have two hypotheses in play here.  One is what’s called Engel’s Law, which is that the expenditure share of food should remain constant when real income remains unchanged, and the other is that we have a good measure of real income.  It’s that second hypothesis that Herb Emery is testing.  Remember that we know what family income levels are because we can count the number of dollars they are receiving and we know what they’re spending on food since we can count the number of dollars they’re spending, and we can observe food’s share of family expenditure because that’s just one dollar figure divided by the other.  So those numbers are pretty solid.  And when we’re calculating real income, deflating nominal income by CPI-Total, we’re using one of our observable numbers again – nominal income.  So Herb’s hypothesis (and the hypothesis of the rest of the literature which uses this approach) is that there’s a problem with the CPI-Total. 

If our calculated real income has remained unchanged but food’s expenditure share has gone up, it must be because real income has actually fallen, and that must mean that CPI-Total is underestimating the increase in the cost of living for our target sub-group.  And if food’s share has gone down, then real income must actually have risen, even though calculated real income has remained unchanged, so CPI-Total must be overestimating the increase in our target group’s cost of living.

There’s a fair bit of statistical analysis involved in the actual research paper, and you can decide for yourselves whether you think Herb has controlled for everything which could have affected the budget share of food, but that’s the general logic behind his National Post piece.


And you thought CPI releases were boring.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.