U.S. Recession Predictor Sends Warning Signal

Includes: SPY
by: Jason Cawley

The simple US recession warning statistical model I published here 2 years ago has just sent a recession warning signal for the first time this cycle.

The current recession risk score is 0.442, above my trading system's recommended 0.40 level "trigger" level to move to a "risk off" asset allocation.

This model is trained to spot late cycle boom conditions one year prior to US GDP recessions, rather than signs of economic weakness from a recession itself.

The present prolonged period of a nearly flat yield curve along with low unemployment are driving the model's prediction.

I expect to remain in risk off positions over the next two years.

In May 2017, I published a simple statistical model trained to forecast US GDP recessions 1 year in advance here on Seeking Alpha. Here is a link to that previous article - A Simple US Recession Predictor

I then explained a market timing system to adjust asset allocations over the cycle in accordance with the statistical model's results that June, and found in backtesting that a 0.40 risk level was as appropriate "warning signal" trigger level to adjust asset allocations by lowering weight in stock and increasing weight in cash and short-term bonds. Here is a link to that analysis - Asset Allocation from Recession Timing

I have since updated the model's readings in a series of articles, most recently one last December when the model approached that warning level for the first time. Here is that most recent article in the series - December Recession Risk Article

I can now report the formal warning signal was sent this Friday morning after the publication of the CPI report for April 2019, which allowed updating of all of the model's inputs. The risk score now stands at 0.442, up from 0.399 as of the April levels of its indicators. Here is an updated graph of the recession risk score time series -

Recession Risk Graph

The blue line in the graph above (my own model) shows the time series of that risk level, and it has doubled in the last 10 months. Below I also give a table of the recent levels of the model's risk score, along with "instant" levels using present indicator values without the model's aligned moving average smoothing.

Those instant levels provide a forecast of where the full model risk level is headed next, and appear as the yellow points in the graph above. The green points, corresponding to the "refitted" values below, report the same using adjusted parameter values for the model based on the data received since it was first published. The blue line and yellow points are based on current data, but the original fit parameters found in May 2017.

Recession Risk Recent Values Table

The recent increase in the risk level stems from the declining unemployment rate and especially the prolonged period with a flat yield curve. The model looks at the yield spread between the 10-year and 2-year Treasury, specifically, as a monthly data point reported by the FRED II economic database from the St. Louis Fed. While that spread hasn't moved much since last November, staying in the 0.15 to 0.20 "barely positive" range, the model uses a 12-month moving average of that indicator, and low current values for that spread are replacing levels near 0.50% from a year ago. That will continue over the next 2-3 months and should cause further increases in the model risk level even if the spread remains unchanged.

In the December article, I thought that process would send the risk-off signal as early as mid-January 2019. The uptick in the unemployment rate to 4.0% reported for December prevented that from happening, and that move kept the risk score just below the 0.40 level throughout the first quarter. The unemployment rate move has since reversed. That "kink" in the rising blue line near the right end of the first graph shows the impact of that brief move higher in the unemployment rate, and is the item I didn't see coming when I wrote the December article. The warning signal has thus sounded 4 months after I expected it, and I get a lesson in waiting for confirmation.

Here is a chart of the model drivers as it actually uses them, with their moving averages and direction of influence. The vertical pink regions show the pattern the model is looking for in these indicators. The raw data here are from the FRED II economic database, with moving averages as used in my model processed in Mathematica -

Recession Indicators Graph

Notice the yield curve moving average touching the 0 line in the past recession and barely above it currently. Also notice the lows in the smoothed unemployment rate series, and how we are currently below the low level of all of the previous cycles in the period shown. The US economy doesn't normally stay at such full employment levels for very long before that rate reverses higher. Keep in mind the pink regions above are defined by being 1 year before US GDP recessions - those upward moves in the unemployment rate moving average basically are the recessions themselves. By the time a 3-month moving average of unemployment is skyrocketing, it is too late for a recession warning to do a trader much good, which is why the model is trained to spot the prior lows, not the increases. There is more uncertainty in that, naturally, but certainty without warning time cannot be traded; everyone knows by then and prices will have adjusted.

It is also worth noticing that the inflation rate is not showing any difficulties. But it is the weakest of the model's indicators, and mostly helps it see recessions even from relatively high levels of unemployment in cases with a very high inflation rate, like the late 1970s and early 1980s. The 2000 recession occurred from levels of trailing CPI inflation around 3%, higher than today but within roughly the same range before and after.

Here is a table of the recent levels of those indicators with their moving averages, to see what numbers are feeding into the model over the recent past.

Recession Driver Levels

Notice that the present level of the unemployment indicator is back to the same value it hit in November 2018, while the yield curve moving average has fallen by half since September 2018. The 3 higher values of the former in the first quarter kept the recession risk score in the high 30s; as soon as it fell back to its previous level, 4 months of decline in the yield curve moving average were "unmasked".

Since rate moves are important to the story here, I also present a graph of the Treasury yields by maturity since the start of 2106 (data from Fred II as before). You can clearly see the Fed's rate increase campaign pulling the low end of the curve upward, with the whole spread compressing as the long rates fail to move higher with them. Then in late 2018, you can see the whole higher end break downward into that snarl of the previously tiered lines on the chart, as we see various "inversions" on the shorter end of the curve.

Treasury Yields by Maturity 1-2016 to 5-2019

The rate curve move is pretty widely known, but the quantity of money story isn't. Last year, the real rate of growth of M2 money supply fell below 1% and stayed in the 1% range or less from May through November of 2018. That was a sharp deceleration from real money growth as high as 6% per year back in 2016, and still around 4% per year in mid-2017. Here is the chart for the real rate of M2 growth since 2000, in order to see the whole past cycle as well as the current one. (Data from Fred II, processed in Mathematica to compose M2 level and CPI price level data into the real year-on-year rate).

Real M2 Growth Rates 1-2000 to 5-2019

The present level of that year-on-year real rate of change is plus 2%, well below its long-term average. Monetary policy must be regarded as tight in a quantity sense, regardless of the low level of rates. The Fed last year was engaged in a major quantitative tightening campaign that essentially undid the third of the "quantitative easing" operations earlier this decade, and brought excess reserves back to the level seen in 2012 after "QE2" was completed. Here is the excess reserves story over the same time span as the real M2 story above, direct from Fred II.

Excess Bank Reserves 1-2000 to 5-2019

Notice that large uptick on the first QE operations corresponding to the sharp spike higher in the real money growth graph, how the pauses between the QEs slowed real money growth, but by QE3 strong real outside money growth had been established. After the QEs end and the Fed's tightening campaign begins, real money growth slows, then drops off sharply, with the Fed's quantitative tightening accelerating the natural decline in excess reserves.

There are signs that the Fed is backing off that campaign since December. The bond market is saying that they overdid it, and recent softness in inflation measures is saying the same thing. Tight money, slow money growth, a flat yield curve, all at a time when we are already nearly 10 years into an expansion that has brought us to multi-generation lows in the unemployment rate, all are saying that this cycle is about over.

It is possible the Fed's reversal might prevent an outright recession and give us a "soft landing". The bond market might be wrong, and long rates might reverse higher and give us a positive sloping yield curve again. If all of that happened, and inflation ticked back up, and the Fed resumed gradual rate increases at the short end of the curve but lagging increases in long-term rates, we might construct a scenario that would avoid recession. But that is not what the statistical indicators are currently saying, or what past history tells us is the way to bet. It is much more likely the Fed was too hard down on the brake for too long last year, and the recession that will cause is already on its way.

I should say a word about what the risk-off signal means for trading. To me it counsels moving at least 40% of asset allocation from "risk-on" assets like common stocks to "risk-off" assets like 1-year bank CDs or 2-year investment grade, Treasury, or municipal notes. Personally, I am willing to move 60% of my allocation in that manner off these timing signals, but it's my model and your mileage may vary.

When does the timing system say to get back in? The answer is that the recession risk score would have to fall to 0.20 or less and remain there for a solid year. These things are not fast tactical allocations, and the likely length of time the trading system linked to the model will have me in a risk-off position is around 2 years, possibly 3. A recession itself could be a full year off, and takes time to occur once it does arrive. This makes 2-year notes a natural vehicle for the risk-off position, though rolling 1-year bank CDs is a reasonable and safe alternative.

Last, I'll say a word about my own timing and trades off the system so far. It was early in December of last year, and even last summer when the risk scores got into the 30s, I began dialing back my own risk level. In hindsight, that was early and I should have trusted the model to be far enough in advance of the event "as designed". In practice, the specific mix of things I sold last year and interest received since, compared to lower dividends, have me within 1% or so of where I would have been had I not moved early. That's within the noise of the last week's stock market moves, and I'm fine with it. I've done very well this cycle and now it is time to keep it.

I hope this is interesting, and as always all questions are welcome.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. 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: I am trading on this system myself and I am currently in my "risk off" asset allocation.