On Estimating the Revisions to GDP Before They're Released

If you've been following my posts, you know that I've often attempted to estimate the revisions to GDP before the release.  Since most of the components of GDP that i described in my earlier post on the revision to 3rd quarter GDP were derived from monthly reports from various agencies that are readily available, I've tried to estimate the impact of each when covering them , or at a minimum, suggest whether the details would result in an addition to or subtraction from the previously published growth figures.  Robert has asked me to evaluate how I've done on those forecasts, which is what this post is about.

The revisions to each GDP report begins to accrue from the week after its release.  For the advance report on the the 3rd quarter, September reports for international trade, construction spending and factories inventories, which were only guessed at by the BEA when reporting advance figures for GDP, were not published until a week later.  In my initial assessment on the revisions to the trade report, i concluded "improvement in our balance of trade should add 0.11 percentage points to 3rd quarter GDP when the 2nd estimate is released at the end of November".   Later i suggested that would be reduced by 3 basis points when the revisions to the import-export price index were released.  As it turned out, i was way off; there was a net revision to trade figures that subtracted 19 basis points from GDP.   My initial assessment was based on chained dollar figures for goods from table 9 in the trade pdf so that should have been accurate.  However, i appeared to have underestimated the impact the revisions to export and import prices, which lowered real exports and increased real import.  In addition, there was a revision to our exports of services from 4.4% growth to 3.7% growth that i had not considered.

On the construction spending report released that same week, we broke it down into its three components, and concluded "while we believe this report will give a boost to the 2nd estimate of 3rd quarter GDP, quantifying that boost with the numbers we have here is next to impossible."  Difficulties in analyzing the construction spending report's impact on GDP stem from the fact that the report covers much less than the BEA includes in the relevant public and private investment components of GDP, and that there are a multitude of privately published price indexes for the various components of non-residential investment that are used as deflators.  Thus, to come up with an estimate of just the real construction component of construction investment, each type of construction would have to be weighted and have it's inflation adjustment computed separately, something we're reluctant to undertake without having a program to crunch the numbers.  Finally, for the factory inventories data released that same week, we concluded "net impact on 3rd quarter inventories for GDP from this report will likely be insignificant."  More on that later..

With the October retail sales report released the next week, we noted the revisions to September and August retail sales and concluded: "the decrease of roughly $0.8 billion in sales for August and September should subtract nearly $3.4 billion from the growth of annualized personal consumption expenditures for goods when the 2nd estimate of third quarter GDP is release at the end of the month, which would hence subtract 0.02 percentage points from 3rd quarter GDP growth"   As we noted when covering the 2nd GDP estimate, "real consumption of durable goods grew at a 6.5% annual rate, which was revised from 6.7% in the advance report', and "Real consumption of nondurable goods by individuals rose at a 4.0% annual rate, revised from the 3.5% increase reported in the 1st estimate", and hence the net contribution from goods to GDP went from 99 basis points to 105 basis points.  THE problems with our estimate in this case appear to be the result of using the aggregate number, and an aggregate deflator, rather than examining each revision by business type, and deflating each business's sale with the change in prices for the type of goods that each business would sell.  Again, to do the detailed work on that would involve taking each revision to each type of retail sales and deflating it with an appropriate price change from the same months's CPI data, which we know from experience is a very tedious chore to undertake manually..

That same week saw the release of the September business inventories report, which adds factory inventories to wholesale and retail inventories to arrive at a figure for total non-farm inventories at month end which should be comparable to the same metric in the GDP report.  However, adjusting those inventories for changes in price, which we'd need to do to get a GDP relevant real inventories figure, is complicated by the various methods different businesses use to value their inventories, and further complicated because the price adjustment is made for the month when the goods enter inventory, rather than for the month demarcated by the report.  Since September inventory data is published two weeks after the initial estimate of GDP, the BEA itself must estimate that for the Advance Report, and as they told us in their technical note accompanying that report, and that they assumed a decrease in nondurable manufacturing inventories, and a decrease in wholesale and retail inventories, without quantifying their assumptions.

So in covering the business inventories report we noted the BEA was wrong on two out of three of its components, and set out to determine the minimal impact those errors would have.  On wholesale inventories, we concluded that "3rd quarter inventories were at least $8.4 billion higher in inflation adjusted dollars than the BEA estimated in their advance estimate of 3rd quarter GDP", which on an annualized basis, would "boost the change in the inventory contribution to the change in GDP by at least $34 billion dollars, or enough to add at a minimum 0.21 percentage points to the next estimate of 3rd quarter GDP growth.  In addition, we concluded that retail inventories "would make for another upward revision of roughly $26 billion at an annual rate to real third quarter inventory growth, which in turn would add a minimum of 0.16 percentage percentage points to the next estimate of 3rd quarter GDP."  Since September factory inventories were down by 0.4%, and the BEA did not quantify their estimate (and we couldn't for the reasons noted above), we did not estimate any revision from factory inventories.  Similarly, since we could not quantify how much the BEA had incorrectly assumed that wholesale and retail inventories had fallen, our analysis was limited to the minimum impact that the actual increase in September inventories would have, ie, if the BEA had estimated a drop of less than 0.1%.  As it turned out, the change in non-farm inventories between the 2nd quarter and the 3rd quarter was revised from an increase of $52.4 billion in the advance estimate to an increase of $85.1 billion in the 2nd estimate, a difference of $32.1 billion, or a quarterly swing in growth in inventories at a $130.1 billion annual rate.  That resulted in a revision to the impact of non-farm inventories on the GDP growth rate from a decrease of 145 basis points in the advance estimate to a decrease of 64 basis point in the 2nd estimate, which we posted on Sunday.  So the error made by the BEA in estimating 3rd quarter inventories was much worse than out minimum case scenario, and hence resulted in a much larger swing in it's contribution to GDP growth than we had anticipated.



good explanation

People usually go by percentage point contribution or annualized percent change. Just FYI but nice explanation.

Filtered HTML

  • Web page addresses and e-mail addresses turn into links automatically.
  • Allowed HTML tags: <b> <blockquote> <br> <caption> <center> <dd> <dl> <dt> <em> <font> <h2> <h3> <h4> <h5> <h6> <hr> <i> <img> <li> <ol> <strong> <sub> <sup> <u> <ul>
  • Lines and paragraphs break automatically.

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.