In August 2011, ECRI declared a recession was upon us. If you spend enough time examining their initial proclamations it was literally that a recession was imminent, saying “It’s either just begun, or it’s right in front of us.” Subsequently they revised their call, saying the recession would hit “by mid year” 2012. At that time an array of proprietary leading indicators were in contagion. Hussman and others followed suit with bearish outlooks.
In January 2012 we said we couldn’t see it. The general response from the bearish camp was we were looking at too many co-incident data and not enough leading data. As the economic data slogged slowly upward with the stock market and the leading indicators, there was an ironic switch in the bearish camp to focusing on co-incident data and even 12-month rate of changes of said data to prop up recession calls. The argument was the normal smoothed growth metrics deployed no longer applied due to massive seasonal structural changes and the extent of the great recession. But the data kept grinding upwards. Leading and coincident. Specific items such as Real-incomes, GDI, jobless claims were cherry-picked in support of the recession call during the course of the year – all the while the stock market was slogging ever upwards and the economy kept muddling along.
More recently, co-incident data was “rolling over”. Doug Short quickly put that one to bed with his “Big Four” updates and this was confirmed with our “NBER Model – Confirmation of last resort” note. Whilst there was weakness in the co-incident data, there was not wholesale “rolling over of the co-incident data” as was being implied. In late June 2012, when some leading data was indeed rolling over, the bears came out to play again. We issued a note in the first week of July 2012 “Here we go again” One observer, already on his second “Recession Right Here, Right Now” call, noted that I was blindsided. We sure were – the S&P-500 has risen 11% since.
Also recently, some observers said that those that were in the bullish camp since September 2011 were “lucky” as they were saved by “Fed Intervention”. Well, I am sorry, but anticipating the Fed actions and factoring them into effects on the economy and the stock market are “part of the game” as far as we are concerned. Ever heard of “Don’t fight the Fed“?
With neither the non-proprietary leading nor the co-incident data currently supporting an overwhelming case for recession, the focus has now switched to “revisions.” The argument goes that the data will be substantially revised down until we suddenly one day wake up to a revision that plunges us in recession at some time in the past. We cannot deny this is not a possibility. But by definition, revisions can go up as well as down and there have been some upward revision of late. It just seems odd to me that one would rely on speculation on revisions to make a recession call. In this case, why even bother with all the amazing leading and coincident models that people have built and focus on? How is it possible that these could be pointing in one direction and the revisions in another? We do not have a quantitative model that predicts revisions. Maybe the others do, but we don’t. For us, factoring in revisions would be mere speculation – and we don’t deal with speculation we deal with the facts as they are presented to us by the various mainstream leading and coincident data we deploy.
On that note, whilst the US economy is experiencing historically sub-par growth (printing new growth low-watermarks) and there are pockets of real worry such as industrial production and real sales, and we acknowledge the risk of external shocks are great to to a weak economy and that revisions could scupper everything – the immediate leading and co-incident indicators are not telling us in a convincing manner that we are headed into recession.
There are just two things of importance when it comes to evaluating performance of recession forecasting. The first is the accuracy of your dates of proclamation versus what eventually gets proclaimed by the NBER. This could take 6-9 months to figure out, given NBER proclaim recessions well after they occurred. The second is how much advance warning you issue. It is very hard to be perfect here but 3-5 months lead is preferable (See “Recession : Just how much warning is useful anyway?“) This is just a measurement of how well you timed the stock market peak. For most fund managers and private investors attempting to manage their portfolios in a risk-adjusted fashion the first criterion is “academic” and the second criterion is all important.
Calls such as “recession within the next 6-9 months” are no good to man or beast. Let us for a moment assume ECRI’s, Husssman, Rosenberg, Shedlock et al initial recession calls (just the initial ones, they have on more than one occasion since then made new “Recession : Right here – right now” calls) were spot-on and we do indeed at some date in the future come to realize that the US fell into recession in July/August 2012. The fact of the matter is that the S&P-500 has risen over 30% since September 2011. When one couples the accuracy of their call (more than one in some cases) with subsequent gains on the stock market right now, it is an “epic fail” for the market participant. The current bullish camp would have to declare recession calls after the stock market had plunged by more than 30% from today to match the performance of the bearish camp on this criterion. Even if the bullish camp only declares a recession after the stock market falls another 20% from today, they will still be ahead of the game on this point.
We do not claim the bullish camps’ models are perfect or even right. But one thing is undeniable – they have kept you on the right side of the stock market until now. Business cycle forecasting is indeed a humbling profession, but when this whole show is said and done all the forecasters will be judged on 3 things – their NBER date accuracy, their stock market peak dating accuracy and how often they cried wolf or changed their story. Depending on what you use recession forecasting information for will dictate what weights you assign to these metrics.