About The Authors

Saturday, July 30, 2016

Predicting SAAR, Deconstructing SAAR

Mike Smitka

As an economist, I avoid the prediction game. I am also wary of reading much into a single month's data. What an economist can do is to provide reasoning why over time a particular average level of sales makes sense, and levels significantly above/below do not.

Let me start with the monthly time horizon. Next Tuesday we'll get the latest sales numbers this coming Tuesday (August 2, 2016). Those numbers will tell use sales down to the last vehicle, except for Tesla, which reported sales of 2,250 for the past 4 months. That's a false level of precision. First, there's human error, though that ought to average out. A sale won't get reported, or digits will get reversed, or ... The Law of Large Numbers though means that while there will be over- and under-reporting by individual dealerships (and DMVs), those will average out and not cause much error in the total for the whole market. But one thing we know is that claiming there were exactly 1,513,086 light vehicles sold in the US in June 2016 is not true. When I present data, I try to round things off to 3 significant digits, here to 1,510,000 units.

I dutifully look at the numbers, but for the next year or so I really don't expect to learn anything from them

Then's there's the conceptual issue, that what "sold" means is less than clear. Dealers face incentives to tweak the numbers to earn "stair-step" incentives where one more vehicle can add a lot to their bottom line. Better to get the bird that's almost in hand by reporting a sale, than to carry it over to the next month when they might fall well short of (or significantly exceed) the threshold with an uncertain payoff. In Europe, discounting takes an indirect form: rather than placing cash on the dash, as it were, a car will be sold and reappear on the dealer's lot as a used car with zero mileage. They're "sold" and they're not. The data again offer a false sense of precision. Again, my sense is that an error (accidental or deliberate) in one month gets averaged out in subsequent months. But it does mean reading too much into one month is inappropriate.

Then there are the random factors, snow storms and holidays that fall midweek and ... there are such every month. Whatever the "true" level of demand, the performance in any reference period will deviate from that. Yes, we can apply seasonal corrections, and try to remember that February sales in a leap year will of course be higher, and sales in a month with 5 Saturdays may also be quite different. Such corrections however are but fancy averages, and so will never get the adjustment quite right.

So what can an economist say? There are the house economists at Ford and the others, who in conjunction with others in management need to provide a number for each product for the coming month and quarter and year for scheduling overtime and shutdowns at the plant level, for issuing purchase orders to suppliers orders for the next 30 days, and for planning capacity. In this role a house economist is as much soothsayer as professional. Formal models get combined with experience to which hunches are added, because at the end of the day there has to be a number. What will the Fed do? Over the past year, much less than the Federal Open Market Committee members themselves had predicted. But even if they bump short-term interest rates by another 25 basis points, will that affect the rates on car loans at all, or otherwise change sales? There's no reason to think they'll get this right any better than the traders who are betting billions on bonds.

why no mention of GDP: some "advanced estimate" components are good, the headline number not

An economist can however put some limits on what is likely to happen, using theory and a reading of the available data (which only show what happened last month or last quarter, never what is happening today, and absent theory tell us nothing useful about what will happen tomorrow). Here I look at two factors that influence automotive sales, interest rates and employment.

Employment first. Over the long run light vehicle sales correlate very closely with total employment, with about 1 sale for every .12 sales for every million workers, With 145 million people employed, that gives a SAAR of 17.4 million. This is not a tight relationship in the short run, and over the full period of the graph shows a downtrend. Indeed, a simple linear regression would suggest that I use .10, though for technical reasons that surely exaggerates the trend. In any case, that hints that 17.4 million is somewhat generous.

Employment continues to increase. Part of that is because the overall population continues to rise. Using age-specific population projections and the relatively stable rates of labor force participation prior to the Great Recession lets me estimate a normal level of employment, the red curve in the graph below. That's rising at about 58,000 a month in mid-2016, falling to about 50,000 by mid-2017 and 26,000 in 2018. In short, fertility plus immigration is barely offsetting the retirement of the Baby Boomers. On that basis the labor force will increase by only 900,000 workers over the next 2 years. That means we won't see SAAR rise by more than 100,000 units, which is smaller than the month-to-month volatility in the sales data.

But as we know, the US economy has yet to fully recover from our Great Recession. Compared against the demographic-corrected trend level, employment remains about 4.3 million below the pre-recession levels. The US economy has shown steady employment growth for the past 5 years, since summer 2011. We've also seen participation rates increase for prime-age workers, though that too remains below pre-recession levels. Barring a distinct slowdown or a boom – nothing in the real estate and residential construction markets suggests either – then we will keep adding jobs for another 2 years. Using the 0.12 figure, that will push car sales up by 0.5 million units. So if I were an optimistic, I could point to a potential SAAR of 17.5 + 0.1 + 0.5 or 18.1 million units by end-2018. I think the likely sustainable sales rate relative to employment is likely closer to 0.11, while the economy faces more headwinds than tailwinds...

How about interest rates? Here the picture is quite clear: they will stay low. First, the Fed is unlikely to raise rates aggressively, given the lack of signs of either inflation or accelerating growth. Second, across the globe growth are down. The developed world, plus China, are aging. The population of Japan is falling in absolute terms, and the working age population is falling in Europe and in China. Then there's productivity: an economy grows even with a fixed number of workers as long as output per worker grows. While we have new gadgets galore, the increase in productivity from having a smart phone is less than that from having a phone. We in the developed world see some gains, but the realignment of work that access anywhen to the cloud enables is only affects a certain share of jobs, is happening only gradually, and is not leading to large gains in output. That example can be repeated for a variety of technologies; see Robert Gordon's The Rise and Fall of American Growth for a systematic analysis. [The work structure example is my own.]

All of this is reflected in interest rates: they have fallen across all maturities, as reflected in bond prices. Furthermore, the yield curve suggests no upturn in interest rates for the foreseeable future (which for US bonds is 30 years), either due to stronger growth or to inflation (or, more accurately, the sum of the two). That strikes me as an odd bet to make at 20 year time horizon, and historically long-term bonds haven't been good indicators of what will happen. In the 3-5 year time horizon, however, the story told by bonds is more credible: we won't see a boom. I have both a basic interest rate graph, and one that looks at the implied yield on 1-year bonds, calculated for example from the difference in 2- and 3-year bond yields.

The final element is energy prices. My track record is abysmal, but so to my knowledge is that of everyone else. (For my posts on energy, See "Another Fracking Saudi Conspiracy Story" and here for "Peak Oil Revisited: Did I Get Anything Right?") From the perspective of extraction costs, the era of really cheap oil is over. For now, however, fracking offers a lot of potential at intermediate prices, while demand growth has slowed and the cost of alternative energy sources has fallen, including both solar and wind. The world has more natural gas than it can consume, and while over time the ability to transport it from where it is produced to where it might be consumed via pipeline and LNG ships will affect that, it's hard to see what might affect prices of gasoline in the US through 2018.

In conclusion, next week we'll see many column inches and blog posts dissecting the latest sales report. At the firm and maybe even segment level, it could contain information, though at the monthly level I'm still reluctant to play that interpretation game. More generally, we'll have more of the same. I dutifully look at the numbers, but for the next year or so I really don't expect to learn anything from them.

Thanks to Dr. Paul Traub of the Federal Reserve Bank of Chicago, Detroit Branch and former head economist of Chrysler for pointing out the strong correlation between employment and sales. This idea can be tweaked in various ways, setting up a multiple regression framework that would incorporate changing vehicle longevity, putting in a separate variable for those employed but over age 65 and for under 25, putting in a variable for changes in gasoline prices, and for the interest rate (or perhaps, using a combination of loan rates and loan maturities and vehicle prices, monthly payments). Obviously I've not done that.