GDPNow forecasts 1.3% for GDP results on Friday.
Part of that forecast includes an adjustment of -1.09 percentage points due to inventories.
For details, please see GDPNow Forecast Dips to 1.3 Percent: One Heck of an Inventory Adjustment.
Since Change in Private Inventory (CIPI) accounts for a whopping 1.09% subtraction in the GDPNow forecast, I would like to understand what's going on.
This is more than a bit complicated if not outright geekish, so apologies in advance.
Total Business Inventories
Inventories keep rising. Note that total inventories (purple) is on a different axis and is smoother than the subcomponents.
Change in Total Business Inventories Q4 vs Q1
Between Q4 and Q1 the change in total inventories is $24.869 billion.
Change in Total Business Inventories March 2019 vs May 2019
Between March and May, the change in total business inventories was $17.657 billion.
Change in Private Inventories by Industry
Those numbers do not remotely come close to my calculations but it's important to note that BEA CIPI numbers are annualized.
To compare to monthly totals one needs to divide by 12. To compare to quarterly numbers, one needs to divide by 4.
CIPI / 4
Now we have monthly and quarterly comparisons in the same ballpark.
But please note the increasing volatility of inventories even when even when recessions are ignored.
With that prelude out of the way. Let's turn to an email discussion.
Email Discussion With Pat Higgins
Mish: Hi Pat
How do you arrive at your initial CIPI?
Note. The answer below very complicated and assumes familiarity with the GDPNow spreadsheet calculations.
Many will want to skip to my next question.
The inventory forecasting model works pretty the same way at the beginning of the forecast cycle as at the end of the forecast cycle except for the fact that there is more monthly data to forecast in the former case. GDPNow forms two independent forecasts of inventory investment One just using the quarterly real GDP subcomponents for the previous 5 quarters ( and a BVAR model). And one that forecasts the yet-to-released monthly inventory source data and builds it up into a total CIPI forecast. The model then uses a weighted average of those forecasts. See cells FV9:FW9 and FV11:FW11 in the tab Inventories in the spreadsheet [the 5th from the rightmost]. At the beginning of the forecast cycle for 2019q2 on April 29, the quarterly BVAR model forecasted $(2012) 44 billion for 2019q2 inventory investment with a weight of 0.49 and the monthly inventory data model forecasted $(2012) 57 billion for 2019q2 inventory investment with a weight of 0.51.
The $(2012) 132 billion CIPI number was the model’s estimate of 2019q1 inventory investment using the published (primarily monthly) data on inventories from BEA and the Census Bureau. The CIPI contribution can be roughly calculated by comparing the 2019q1 CIPI number with the 2019q2 number. If the 2019q2 CIPI number is above the 2019q1 CIPI number, the CIPI contribution will be positive. If the 2019q2 number is below the 2019q1 number, the CIPI contribution will be negative.
The 132 billion should be most similar to the line 1 number in NIPA table 5.7.6B (which has real data) not the line 1 number in NIPA table 5.7.5B (which has nominal data). The 132 number would be exactly equal to line 1 in table 5.7.6B if the BEA was not going to revise 2019q1 CIPI before or at the time of the advance 2019q2 release. Since the 2019q1 data will be revised with the benchmark GDP revision on July 26, 2019; GDPNow does not force this equality. I.e. GDPNow tries to anticipate the revision to 2019q1 CIPI since the CIPI contribution to GDP growth depends on 2019q1 CIPI. Whenever the monthly wholesale/retail/manufacturing inventory data from the Census Bureau, the GDPNow estimate of the previous CIPI estimate can be revised as long as the previous quarter CIPI data will be revised at or before the advance GDP release for the upcoming quarter. You can see how the 2019q1 CIPI number is arrived at by going to cell FS13 in the Inventory tab in the spreadsheet, and referring to all the cells that it is linked to [there are many of them, including columns FX and FY ].
Monthly data on retail sales, manufacturing shipments, etc., are used by the GDPNow model to forecast CIPI [see tables A8a – A8c on pages 77 – 84 of the GDPNow working paper https://www.frbatlanta.org/-/media/documents/research/publications/wp/2014/wp1407.pdf ]. So some of the same monthly data used to forecast consumption, business equipment investment and other GDP subcomponents is used to forecast CIPI. But beyond this commonality in data sources, there are no enforced consistency conditions between how the model forecasts CIPI and how it forecasts any of the other GDP subcomponents.
Mish: [Note - This was a follow-up that included some of the above charts].
In my extremely crude example, assuming I am supposed to divide by 4 and assuming CIPI drops from 30.4 to 18.7 is there crude way (inside or outside your spreadsheet) to calculate CIPI contribution to GDP (again assuming no further changes in the final CIPI estimate of 18.7)?
Is there a page on your spreadsheet where I can say CIPI = 18.7 and crank out a net contribution total?
I can’t really say whether your approximation is good or not because of the FOMC blackout. But I can say that one ingredient your approximation omits is the inventory valuation adjustment (IVA). See line 1 in NIPA underlying detail table 5.7.5BU3. Your calculation will approximate the change in the book value – see line 1 of table 5.7.5BU2. (Nominal) CIPI is the change in the book value of inventories plus the IVA. Sometimes the IVA is important. For your approximation, where I take it you are assuming June inventories = May inventories, you’d want to multiply the difference between May and March inventories by 4 as in cells FY21:FZ21 of the Inventories tab of the GDPNow spreadsheet.
It looks like there have been only 8 quarters since 1948 when the CIPI contribution was negative and the change in real CIPI was positive or the CIPI contribution was positive and the change in real CIPI was negative. The BEA uses something like a Fisher chain-weighting formulas to calculate the contributions [see ] that can get a bit complicated with a lot of subcomponents [GDPNow uses a similar calculation, but the GDP subcomponents are less disaggregated than what the BEA uses]. I’m not sure if the BEA only uses farm and nonfarm inventories in their contribution calculation or if they use even more detailed subcomponents (you’d probably want to contact them).
The contribution calculation is not done in the spreadsheet because the formula is fairly complicated (see pages 37-38 of ).
Often it will be the case, that you can approximate the contribution by multiplying the CIPI contribution for a quarter in the recent past, by ratio of the change in the real CIPI in the current quarter to the change in the real CIPI for the quarter in the recent past you are using. You may want to experiment to see if you can approximate the 2019q1 CIPI contribution that way.
End of Emails
Yes, the above discussion is geekish.
Unfortunately, even ignoring the Inventory Valuation Adjustment (IVA) I am not positive on my approximation as he could not comment due to the blackout.
The point of this post then is simply to highlight not only the volatility of the CIPI but the complicated nature of it.
I believe this is the key component to his reply: "Often it will be the case, that you can approximate the contribution by multiplying the CIPI contribution for a quarter in the recent past, by ratio of the change in the real CIPI in the current quarter to the change in the real CIPI for the quarter in the recent past you are using."
That is as complicated as I care to make things.
I will try again next quarter after I see how this plays out this quarter.
This quarter is a real crapshoot, even more so than normal.
Not only have the past two months' data been extremely volatile, so are the possibilities of existing inventory valuations.
Take cars for instance. Inventories on dealer lots are sky high. Those will have to be discounted distinct from normal inventory fluctuation with no price changes.
I believe GDPNow will be close to the mark. But I also expect offsetting errors.
With no confidence in my estimate, I will take a stab at Real GDP of 1.4% but with CIPI at something like -0.40 and real final sales at 1.8% vs the GDPNow estimate of 1.3% and 2.4%.
We find out shortly.
Thanks again to Pat Higgins and for his generous time in responding to my emails.
Mike "Mish" Shedlock