I've just updated Project Link, my attempt to piece together the bits and pieces of published Statistics Canada data into a coherent history of the post-war Canadian economy. (The introductory post is here, and a post on the first update is here.)

Here are the main elements:

- Incorporating Statistics Canada's long-awaited publication of its expenditure accounts from 1961-1980. As you may recall, when they revised the GDP numbers back in 2012, the new series started in 1981 instead of 1961. (This is what I had in mind when I fumed that "Statistics Canada must be the statistical only agency where the average length of its time series gets shorter from one year to the next.")
- Pushing back the starting date for the LFS and SEPH employment numbers to 1946, along with unemployment rates and average weekly wages. The CPI and CPI inflation have also been pushed pack to 1946.
- Including the income measures of the economic accounts : GNI (what we used to call GNP), NNI and disposable income. I've also gone out on a limb and tried to construct measures of Gross Domestic Income consistent with StatsCan's methodology, and that's covered in a separate post.

More details below the fold.

Statistics Canada's long-awaited (I had given up hope) publication of expenditure accounts from 1961 to 1980 provides a real-time test of the assumption that data redefinitions affected only the level of a series, and not its growth rate. To no-one's surprise, the updated StatsCan data for 1961-1980 are different from the estimates I posted earlier. Here are some summary statistics of the revisions:

Nominal | Real | |||||
---|---|---|---|---|---|---|

Mean | RMSE | 1961Q1 | Mean | RMSE | 1961Q1 | |

GDP |
0.02 | 0.41 | -0.74 | 4.10 | 5.00 | 6.58 |

Consumption |
0.20 | 0.52 | 0.84 | 6.16 | 6.86 | 9.99 |

Investment |
-4.67 | 4.80 | -5.13 | 8.34 | 8.73 | 4.90 |

Government |
-0.30 | 1.34 | -4.69 | 0.94 | 1.54 | -3.18 |

Exports |
0.76 | 1.13 | 0.93 | 16.99 | 26.30 | 20.60 |

Imports |
-2.31 | 2.48 | -1.68 | 17.98 | 27.09 | 30.48 |

I've included the revision to the 1961Q1 observations, because all of the 1947-1960 observations are adjusted in proportion with the 1961Q1 adjustment in order to keep the same growth rates. I take comfort in noting that for the nominal series - and especially for nominal GDP - the revisions aren't egregiously large, but the real series saw some substantial upward revisions.

It would appear that the issue here is the well-known upward bias for inflation produced by fixed-basket price indices such as the CPI. The data before 1961 use a constant 1986 basket to calculate price indices that are in turn used to calculate constant-dollar estimates. It is well known that the use of fixed weights induces an upward bias; this is why Statistics Canada now uses a Fisher chain methodology in which the change in the price index is the geometric mean of the changes in the Laspeyres and Paasche indices.

The further away you go from the base year, the more of a chance you'd expect to see larger changes in relative prices, which in turn induces larger upward biases in the price index. This pattern is borne out in the revisions; increasing as you go back from 1961. Of course, overestimating the price index is the same thing as underestimating the volume estimate, so this shows up in upward revisions for the real estimates. The biggest revisions are for exports and imports, presumably because their composition would be highly sensitive to changes in the relative prices of tradeables.

I can only note the problem at this point, and hope that the relative price changes between 1947 and 1960 were not too important. If anyone knows of a way to modify the Laspeyres index so that it more closely matches the Fisher index, I will be forever grateful.

Moving on to the monthly coincident indicators, the SEPH estimates for employment and earnings have been pushed back to 1946. I've also revisited the CPI estimates. One of the problems with the original series is that the current base is 2002=100, so the estimates in the late 1940s are somewhere around 9.8, and only presented to one decimal point. This leads to some very choppy estimates for inflation in which the dynamics are mainly driven by rounding. So I used the month-to-month variations in the old Cost of Living Index from that time to make the index less granular; the new CPI series are presented to two decimal places up until 1968.

The Labour Force Survey started as a quarterly exercise in November 1945, and I've seasonally adjusted (using the 1953 adjustment factors published by Statistics Canada) and then interpolated monthly numbers from the quarterly estimates.

Oh yes - in the interval since the last update, Statistics Canada terminated two of the 10 coincident indicators that I had been tracking (manufacturing and retail sales) and started new series. The old data have now been spliced to the series that is currently being updated by Statistics Canada.

Here are some charts I've put together. First up is the real weekly earnings series. In its previous incarnation, it looked as though the increase in real wages might have started before 1950, but it turns out that the wobble around 1950-1951 was in fact an inflection point; real wages went sideways in the years immediately following the war:

Here is the unemployment rate:

The bursts of inflation in the late 1940s were a surprise to me:

And finally, employment growth rates. It turns out that employment growth in the 1950s and 1960s was not as uniformly strong as popular nostalgia would have us believe:

And just to keep things interesting, CANSIM is no longer a thing - this is from Bill Joyce at Statistics Canada "We will not continue the branding of Summary Tables or CANSIM, but all of this content will still be available. Users familiar with CANSIM will still be able to search for the previous table numbers and the search by vector functionality will be maintained."

It'll be interesting to see how the new data portal at https://www150.statcan.gc.ca/n1/en/type/data?MM=1 is to work with, and if it links stuff together better than CANSIM did.

Posted by: Frances Woolley | June 04, 2018 at 03:52 PM

Well Frances, it will be interesting. It has become increasingly difficult to put long term time series for Canada together from Statistics Canada data and we will have to see if this changes makes things better or worse. I am not too optimistic but would look forward to being pleasantly surprised.

Posted by: Livio Di Matteo | June 04, 2018 at 06:51 PM