A few months ago, Bank of Canada Governor Stephen Poloz announced he was giving up on economic models as the Consumer Price Index (CPI) stubbornly remained at or below the Bank’s target minimum rate. Then suddenly the CPI rose and remained at or above the Bank’s target midpoint as other economic indicators continued to show slack. Last week Mr. Poloz attributed the recent price surge to “temporary effects” while announcing the Bank was exploring a ‘neutral interest rate’ policy, effectively abandoning inflation targeting.
A closer look at how the CPI is produced may shed some light on why the measure has confounded the Bank of late.
The CPI is based largely on two surveys, the CPI price survey and the Survey of Household Spending (SHS); both in turn rely on other Statistics Canada surveys. (Over)simplified, the expenditure survey is supposed to capture the composition and quantity of products and services Canadian households consume over the course of a year, while the price survey is supposed to capture the prices and price changes for those products and services.
While most StatsCan surveys employ what’s called probability sampling, the price survey, with rare exceptions, uses a form of non-probability (‘judgmental’) sampling. The selection of geographic areas, retail outlets in those areas as well as products / services sampled from those outlets is arbitrary, making it impossible to either estimate sampling variability or identify possible bias. (While StatsCan notes retail outlets are selected based on high sales revenues, cut-off sampling is not used.)
StatsCan’s Industrial Producer Price Index (IPPI) has used probability sampling since its inception in the 1980’s. Why the agency continues to use non-probability sampling in the CPI is unexplained.
The exceptions to the price survey’s ‘judgmental’ sampling are rents, hotel and motel accommodation, and textbooks.
Rent quotes are based on a special module attached to the Labour Force Survey (LFS). Problem is LFS respondents are cycled into the survey sample for six consecutive months at a time. This means rent changes for LFS respondents who happen to move outside the five month window (following their initial screening into the survey) are never captured. Since those who move generally do so once over the course of a year, many dwellings in the LFS sample never record a rent change. Comparisons of rent change with CMHC rent data suggest that CPI rents show too little price change.
The CPI basket weights are based on household spending patterns from SHS. While SHS uses probability sampling, its small sample size, questionable methodology and poor response rate produce results that do not accurately represent all Canadian households.
The SHS collects both household income and expenditure information, used to determine spending patterns for different income levels. However, capital gains are excluded from household investment income.
Compounding that problem is a ‘quality control tool’ that excludes households with a difference between receipts (income) and disbursements (expenditure plus net cash flow) greater than 30%. (User Guide for the Survey of Household Spending, 2009)
This means households generating a significant share of their income from capital ownership – a popular topic these days – are more likely to be excluded from the SHS by design. (Since expenditure from capital gains is necessarily included while the income itself is excluded.)
Not that such households needed help being excluded from the SHS; the survey is voluntary. [Its predecessor, the Family Expenditure Survey, was mandatory.] Current CPI basket weights are based on the 2011 SHS. Only about 2 of 5 sampled Canadian households responded to the survey in 2011, down from about 2 of 3 in 2009.
(Aside: StatsCan stopped producing public use microdata files (PUMFs), for SHS after 2009. For context, the US and UK household spending survey PUMFs have been updated through 2012.)
In addition to the high-income households excluded from SHS by design, a significantly greater share of households in ‘high-income strata’ simply don’t respond to the survey. (Survey of Household Spending 2009: Data Quality Indicators)
That’s a problem. As we’ve learned from Capgemini and other sources, Canadian high net worth individuals (HNWI) – the ones most likely to generate a significant share of their income from capital ownership – faired quite well following the Great Recession. The rest of the population, not so much.
While HNWI rising tides may not have lifted all boats, they would have lifted average household spending – and, by extension, average consumer prices. That is if their spending was accurately reflected in the stats, which it wasn’t.
Bottom line: While the CPI may be confounding the Bank of Canada Governor, an effort at fixing the StatsCan tools at the Bank’s disposal may be more worthwhile than abandoning the inflation indicator that’s supposed to guide its policy for anecdotal evidence and regular meetings with chief executives.
That is assuming the objective is well-informed policy. It would be unfortunate if the Bank of Canada is choosing to go the low-information route that the Government of Canada seems intent on travelling these days.