The Drawing Board
The Model (shown in red) has been substantially revised. The intention behind the model is to find the 7 factors that work the best together over the last 11 years to forecast quarterly GDP. After two consecutive big forecast misses and several factors in the model with deteriorating usefulness much work was done to shore up the model. S&P 500 earnings, the concurrent price of oil and the 10 year Treasury yield were dropped. In their place are the S&P 500, Cass Freight Index and real private inventories.
Evaluating factors for a model is a combination of art, science and guesswork. Each factor is statistically evaluated for the signal to noise ratio (t test), its interaction with the other factors and how its correlation in the first half of the data compares with the second half. If the correlation in the second half of the data starts to weaken or the signal to noise ratio falls the factor becomes a candidate to be dropped from the model.
The current version of the model has the highest overall correlation with quarterly GDP I have had, but it is too soon to get excited. It could be that I have found factors that correlate in the 11 years by coincident rather than as an actual influence or accurate indicator. As Ronald Coase the Nobel laureate noted, "If you torture the data it will confess." I will be closely watching the data on the new variables. If the correlation is only by coincident it should become obvious within two or three quarters.
The Seven Factors
The S&P 500 (purple line) is widely considered a leading indicator for the economy. In the last 11 years it is more of a coincident indicator. When evaluated alone against quarterly GDP it had a two weak lead. When evaluated in the context of the other variables in the model it had a two week lag. The last segment in the purple line shows the average price of the S&P 500 for the 13 weeks from the third week of October through the second week of January was 7.6% higher on an annualized basis than the average price in the prior 13 weeks. This is a weaker gain than the third quarter had. Shifting the lead time by a week either direction lowered the correlation.
The third week in January had daily closing prices averaging 2268.69. This is up 12% (annualized rate) from the average of the 13 weeks associated with Q4. You should be able to see a small light purple dot representing this week just to the right of the pink axis and the last purple line segment. So with 1/13 th of data in, this variable suggests Q1 will be slightly stronger than Q4 2016. Based on the signal to noise ratio this is the 6 th most important factor in the model.
Real retail sales (Orange Line) grew at an annualized rate of 3.2% in Q4, a bit stronger than in Q3. With shuffling the model this factor's correlation maximizing lead time dropped to concurrent from leading 1 month. Retail sales ranks 4th among the factors in the accuracy ratio.
Cass Freight index from Cass Information Systems, Inc. (Black line) measures the volume freight activity in North America. Its correlation to quarterly GDP maximizes with a 2 month lead time. The months associated with Q4 grew at an annualized rate of 4.3% from the averaged of the months associated with Q3. By associated I mean adjusted for the 2 month lead time. The index ranks 3rd out of 7 in the accuracy ratio.
The index is not seasonally adjusted you can see the pattern that it consistently shrinks in Q1. I attempted to construct my own seasonal adjustment for the index, but the results did not have as good a correlation in the model as the index itself. The conclusion I am leaning toward is that the other factors in the model are adjusted more for seasonality than GDP itself and that the seasonal nature of this index helps correct that gap in adjustment for the model. Consistent with the seasonal nature, the data points for November and December (last two grey dots on the chart) suggest Q1 2017 GDP will weaken.
Industrial production (green line and dots) suggest an annualized Q4 growth rate of 0.8% down slightly from Q3. The lead time in the model held steady through revisions at 2 months. Together November and December data points suggest Q1 will show a slight decline, despite the very strong December number. However, if January comes in strong the outlook could be different and of course there are often significant revisions in the prior two months when a new data point is released. This factor remains the most important and accurate in the 7 factor model.
Housing starts (brown line and dots) indicates a surge in growth for Q4. There were 0.16 more housing starts per thousand population than during Q3. This is the weakest of the 7 factors. The scale on its axis (like all the others) is adjusted to show the actual input of the factor to the model. You probably notice the fluctuations of the brown line are smaller than the other factors.
Real private inventory (blue line) implies Q4 will weaken and that the economy will continue weakening into Q3 2017. This is the measure of inventory used in calculating GDP and is new to the model. It has the 2nd strongest signal to noise ratio and a 5 quarter lead time with a positive correlation. However, it is intuitively suspicious.
When I was testing it for the model I was expecting to find a negative correlation where strong inventory growth one quarter would correspond with weaker growth a few quarters out. In testing inventory directly against GDP the strongest correlation was concurrent and positive which is as it should be since growing inventories is part of what is measured in GDP. The correlation turned slightly negative with a 3 quarter lead time flipped back to slightly positive at 4 quarters. Then it got progressively more negative with the strongest correlation at a 9 quarter lead. Implying strong inventory growth is bad for the economy two and a quarter years later. While that's a bigger lead time than I expected, it makes intuitive sense.
However, when tested in the context of the other factors, the strongest correlation was positive with a 5 quarter lead. It was even stronger than the concurrent correlation. The nine quarter lead correlation was almost nothing. At this point it is not clear whether the strong statistical readings of this factor are a fluke coincidence of the last eleven years resulting from torturing (over fitting) the data or a meaningful influence on growth. Since it is in the model I have given it the benefit of the doubt, but will be monitoring its statistics closely with each new quarter of data.
Oil (Gold line and dots) rose over 19% or 102% at an annualized rate 21 months ago. The correlation suggests this will hurt growth in Q1 2017. The 21 month lead time remained consistent throughout the model's revision. Oil had been increasingly constructive to growth in the 2nd 3rd and 4th quarters of 2016. This is the 5th ranked indicator in the model. Oil as a positive concurrent indicator (which used to be the black line) dropped out of the model as mentioned earlier.
Q3 recap: GDP stands at 3.5% growth up from the initial estimate of 3.2%. The model had forecast a much weaker 2.3% which was between the then GDP now and Blue Chip consensus. The significant gaps between the forecast and actual data for Q2 and Q3 sunk statistics for several factors in the model and led to replacements.
2017 will likely start off with a weak first quarter. Two factors show the economy weakening two more have two thirds of the data that collectively show a weaker Q1. So 4 out of 7 factors point to weakness. If the data for the other factors comes out weak in coming months Q1 could even be negative and the start of a recession. At this point it does not appear reasonable to forecast recession in Q1. However, this expansion is long in the tooth. If it survives 3 more months it will displace the Reagan expansion as the third longest in US history.
The current expansion is running at 90 months, presuming a recession did not start in January. If the economy grows through April it will displace the Reagan expansion (92 months) as the third longest in US history. Reagan's, of course, will have had stronger growth.
The last four Republican presidents to get reelected had a recession start in their first year in office. The advantages to this are that it is quite plausible to blame it on your predecessor and the economy will likely be growing strongly as you run for reelection. George H.W. Bush didn't get a recession until a year and a half into his term and the economy was not yet on firm footing as he ran for reelection.
Presidents Kennedy/Johnson and Obama were aided by recessions that started under their predecessor that spilled over a few months into their term. Carter had three strong years of growth and then a short 6 month, but deep recession as he was running for reelection. Clinton remains the only president with no months of recession.
A recession starting soon would fit well into President Trump's narrative of "American carnage" and likely benefit his reelection.
Our 5 year model (red line below) suggests the economy should be faltering. The gap between the actual 5 year growth rate (black line) and the forecast now exceeds the level preceding the great recession. This could mean the model is going bad, the data is over fitted or that a very sharp recession is pending. The next few quarters should tell the tale.
Market implications Bad news will be Bad news
Quarterly stock market (NYSEARCA:SPY) returns should weaken further in Q1 along with economic growth and could be slightly negative. Until the market tries and fails at making a new high a couple of times there is little chance of a sharp decline. The current high was on January 6th and we are close enough to that high that new highs are still quite possible. At this late stage in the business cycle and with the stock market running roughly coincident to the economy or perhaps lagging slightly, bad news in the economy will likely be bad news in the market. The market is not priced for a recession.
Disclosure: I am/we are short SPY.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Additional disclosure: There is no guarantee analysis of historical data their trends and correlations enable accurate forecasts. The data presented is from sources believed to be reliable, but its accuracy cannot be guaranteed. Past performance does not indicate future results. This is not a recommendation to buy or sell specific securities. This is not an offer to manage money.