Almost one year ago, in March 2015, we explained how "The Fed's Artificial Steepening Of The Yield Curve" has resulted in many unexpected consequences, the most important of which has been the erroneous interpretation of the yield curve as a leading recessionary signal. As said back then, "the artificially steep yield curve is a reflection of policy intent not economic reality.... Where the yield curve in the all-important belly of the 5s10s might have deeply inverted in the past just prior to recession, there is no justification to expect the same attainment of absolute levels where artificial monetary intrusion has pushed the curve much, much steeper."
One week ago, it was as if a light bulb went off over Wall Street's head, when first Deutsche Bank's Dominic Konstam realized the significance of the above excerpt, and admitted that far from the 4% recession odds that the Fed's hopeless FRB/US DSGE computer model spews out when looking at the "normal" yield curve, when normalizing for the Fed's intervention odds of a recession in the next 12 months soar to 50%!
In a special report published earlier this week, we noted that today’s near-zero interest rate regime does not allow the yield curve to freely invert or even flatten too much because of certain structural limits. For example, liabilities-driven investors who in the past could receive long rates below the fed funds rate can no longer do so once rates are floored at zero. Investment fund managers are also restricted by mandates from buying negative yielding assets that lead to mark-to-market losses on their portfolios. Pension investors, who must target returns based on liability assumptions, have been driven into high yielding non-core rate assets as their discount rates are stubbornly and unrealistically high compared to Treasury yields. These factors keep the curve artificially steep even though both short and long rates have been clearly trending downward over the years.
An "artificially steep yield curve" - almost as if that's exactly the phrase we used before. What DB did then is the logical next step: to adjust the artificial yield curve and exclude the Fed's intervention.
To address the artificial steepness of the curve we corrected the 3m10y spread for the level of the rates. Specifically we regressed the spread against the short rate, leaving the residual which by definition removes for the bias of the rate level and is centered at zero. Using this new curve as model input, we found the probability of a recession in the next 12 months is 46 percent, considerably higher than the original Fed model has predicted.
But wait, there's more, because while the short-end remains anchored, with every 25 bps tightening in the 10Y yield, recession odds rise by another 6%.
As it may be useful for investors, we attempt to handicap the relationship between the yield curve and future recessions captured in our model. Holding the 3m rate constant, every 25 bps rally in 10s (implying an equal flattening in 3m10y) raises the recession probability by 6 percent. If 10yr yields rally to 1.50%, our model predicts a 59 percent chance of recession in the next 12 months; at 1.00% 10s, the probability is 71 percent.
At Friday's close, recession odds are well over 50% according to DB's model.
Or perhaps far higher, because shortly after DB admitted what we said in early 2015, namely that everyone who was looking at the yield curve as is was wrong, Bank of America's Ruslan Bikbov did the exact same analysis and ended up with a far more disturbing conclusion:
The US Treasury curve is still steep by historical standards. Taken at face value, this may suggest recession odds are small. However, we argue this logic is flawed because the curve is structurally steep when the Fed Funds rate is close to zero. When adjusted for the proximity of rates to zero, the curve may already be inverted and therefore may already be priced for a recession.
And numerically: "Implied recession odds are as high as 64% if the adjusted OIS curve is used"
Laughably, this comes from the same bank whose chief economist Ethan Harris recently "predicted" US GDP for the next decade and forecast there will be no recession until 2027... the same Ethan Harris as profiled in "Perma-bears" 1 - BofA Economist 0."
* * *
Below are the full wonkish details from BofA for all those Wall Street strategists who still hold on to the erroneous creed that recession odds are non-existent if simply looking at the unadjusted yield curve.
A leading recession indicator
We received numerous questions on the shape of the US yield curve and its relationship to recession odds. With the sharp weakening of US manufacturing data in recent months, recession risks are on everybody’s mind, while the curve has the reputation of one of the most powerful leading recession indicators. The basic fact is likely well known to our clients: each US recession since the mid 1950s (when Treasury bond data become available) was preceded by an inverted or extremely flat curve within one year before a recession start (Chart 8). This is, of course, intuitive because a flat curve reflects lower growth and/or inflation expectations. Some of our clients and market commentators pointed to this fact and the relative steepness of the curve to argue that current recession risks are rather low. Indeed, the 3m10s curve at 155bp is still far from being flat (Chart 8).
Mind the zero bound
However, we believe that a simple mechanical extrapolation of the past link between the curve and recession odds is flawed. In particular, curve-based models calibrated to pre-2009 data are likely to underestimate recession odds today. This is because with the Fed Funds at only 38bp risks for plausible Fed Funds paths are asymmetric. Although tighter policy paths are unconstrained, the room for further cuts is likely limited resulting in a steepening bias for the curve. To see this, imagine an extreme situation where the policy rate is at zero and negative rates are not feasible. In such a scenario the curve simply cannot invert, and must be necessarily biased steeper relative to its historical distribution.
Granted, we cannot rule out the possibility of negative rates in the US, but it is safe to say the Fed’s reaction function must be highly asymmetric around zero. Because negative rates entail significant risks for the financial stability of money market funds and the banking sector, the negative growth/inflation shock required for a cut below zero should be larger than positive shocks required for a hike of a comparable size. In addition, negative rates should be floored by the storage cost of currency, another reason for asymmetric risks around zero. In any case, the market currently sees only a small chance of negative rates in the US (Chart 9). The end result is structural steepness of the curve at near-zero Fed Funds levels.
Don’t wait for the curve to invert
Even a casual look at other countries with near-zero policy rates confirms that the curve does not need to flatten significantly for a recession to occur. Consider Japan, a country with the longest zero-rate history. Japan had a recession in 1991-1993, which was preceded by an inverted curve, consistent with past US experience (Chart 10). But note the call rate was at 8% when the curve inverted. Since 1995, the call rate has not exceeded 50bp. Over that period, Japan had four official (announced by the Committee for Business Cycle Indicators) recessions, none of which was preceded by an inverted curve (Chart 10).
A number of other G10 countries adopted near-zero rate policy regimes since 2008 and experienced recessions since then. Some of these recession episodes are analyzed in Table 3. We use two methods to identify recessions: technical definition (at least two consecutive quarters of negative growth) and recession dates reported by Economic Cycle Research Institute (ECRI), which employs methodology similar to that of the NBER in the US. For each recession episode, we report the range of the 3m10s government curve and the policy rate observed for a year immediately before the beginning of a recession.
Again, an inverted curve did not emerge to signal an imminent recession. In fact, in some cases the curve ahead of recessions was steeper than in the US today. As an illustration, Chart 11 shows the historical German curve. Consistent with typical US experience, the curve flattened to extreme levels ahead of each of the pre-2009 recessions. However, Germany also had a technical recession in Q4 2012-Q1 2013 when the ECB rate depo rate was at zero. Not surprisingly, the curve remained steep before that recession. In fact, the curve did not flatten below 120bp in the one-year period ahead of the recession.
Adjusting the curve for zero-rate effects
Although the curve cannot be taken at face value in a near-zero rate regime, we believe it may still provide useful information about recession odds if adjusted for the zero bound effect. The idea is to estimate a model-implied curve that could be prevalent today if negative rates were just as feasible as positive. The curve adjusted in such a way may be directly compared to its historical distribution. As a result, it may be a better recession signal than the observed curve.
Turning to technical details, we model forward rates with a truncated (at zero) normal distribution, calibrated by matching its mean and standard deviation to forward rates and at-the-money option prices. We then compute adjusted forwards as the mean of the corresponding distribution without truncation (hence, using a symmetric distribution around the mean and allowing for negative rates). Although the choice of the truncated normal distribution is somewhat arbitrary, it provides a simple tool to model the core of our argument. Because very long-dated options are not liquid, we analyze 3m5s rather than 3m10s (normally used in academic literature) Treasury curve for this analysis. We found only a small deterioration in R2 statistics for recession forecasting probit models when the 3m5s curve is used instead of 3m10s. Consistent with intuition, the 3m5s curve adjusted in such a way has been significantly flatter than actually observed curve (Chart 12).
Technical factors contributed to Treasury curve steepness
Further, the Treasury curve may be currently skewed steeper by technical factors. Treasury bonds in the belly of the curve dramatically cheapened in the past few months, which is evident from extremely tight levels of swap and OIS/Treasury spreads. As a result, the Treasury curve now looks very steep to OIS. While the 3m5s Treasury curve is at 92bp, the corresponding OIS curve is only at 56bp (Chart 13). The likely reason for this is reserve selling of foreign central banks who need to support national currencies against the recent USD appreciation. International reserves of world central banks declined by about $1tn since September 2015. At the same time, the ability of dealers to absorb the supply has declined in recent years due to regulatory pressures on balance sheets.
Conventionally, academic literature on recession forecasting uses Treasury curve data. But the Treasury curve may not be the best measure of market expectations, presumably the key component of the curve predictive power. Because of the technical nature of the recent Treasury cheapening, the OIS curve should be a better measure of market expectations, and therefore may be more relevant for assessing recession risks.
The curve may be priced for a recession
Applying our methodology to the OIS curve, we found that the adjusted 3m5s OIS curve at -30bp is already inverted. This suggests that the curve already could be priced for a recession (Chart 12). Granted, our methodology signaled a false alarm in 2012 when the curve was also inverted but a recession did not follow (Chart 12). However, at that time the curve flattened to extreme levels because of the forward guidance, an unprecedented event in the history of US monetary policy. In contrast, this time the curve flattened following the Fed hike, which looks more like a typical curve inversion episode. In fact, the Fed was hiking in all previous historical episodes where the curve inverted ahead of US recessions (Chart 8). From this point of view, the current curve flattening may be more worrisome.
Implied recession odds
Our economics team sees only about a 20% probability of a recession in the next year. They argue that the two most important causal factors in recession--aggressive Fed tightening in a battle against above-target inflation and very high oil prices--are not evident today. They also argue that both "real" and financial bubbles are small. The only sector that overexpanded in the recovery is the tiny oil and gas sector (about 2% of the economy at the peak) and the high yield sector overshot fundamentals, but it is much less important than the housing and equity market bubbles of the last two cycles.
Nonetheless, clearly markets are worried and an indicator we have developed confirms their concerns. To quantify implications from the inversion of the adjusted curve, we follow academic literature to compute model-implied recession probabilities from a standard probit regression based on the curve. We acknowledge this type of a model is highly simplistic and does not take into account all the complexities of today economic environment. Still, model probabilities may be interesting to know given the curve’s track record.
We estimated a standard probit model to pre-2009 sample when zero rates were not an issue. We then computed implied probability of a recession within next 12 months with different assumptions about the proper curve to be used in the current regime (Table 4). The model implies about 32% recession odds if the Treasury curve is taken at face value. Just using OIS instead of Treasury rates brings this probability to about 42%. Implied recession odds are as high as 64% if the adjusted OIS curve is used (Table 4).