Now that moral hazard has been adopted everywhere, and the fate of the entire western world is determined by the successful issuance of hundreds of billions of dollars each and every month (we have gotten to the Maginot line where even a hint of a failed US auction would immediately blow up the global capital markets), it is prudent to take a detailed look into a topic that few have covered previously, namely what does the auction demand curve imply. We refer to the distribution of the Low-Mid-High yield break points in each and every treasury auction and whether they can provide some addition insight into the demand picture behind US sovereign debt.
We present the interesting thoughts of Goldman's Michael Vaknin and Anna Stupnytska on this little discussed topic:
Bond Auctions: Shaping the Demand Curve
In this Focus piece we take a closer look at auction data in the US Treasury market and suggest a dimension that is frequently overlooked by market participants—the shape of the auction demand curve. Our empirical investigation suggests that for 10-yr and 30-yr US Treasury auctions, this information is useful for gauging market sentiment and has some predictive power for bond returns in the days following the tender. For the 30-yr sector, in particular, the shape of the demand curve appears to have superior information over other commonly watched metrics.
US Treasury auctions have been closely monitored by investors as a way of gauging the market’s ability to digest additional supply. All else equal, a lower-than-expected auction yield and/or a higher-than-expected bid-to-
cover ratio are seen as encouraging signs, and are often associated with a rally in the secondary market. The
opposite occurs when auctions reveal a higher yield than that prevailing before the tender (a ‘tail’) and/or low
demand relative to the auction size. Here we focus on a particular auction dimension that is frequently overlooked by market participants—the shape of the auction demand curve.
We find that the shape of the auction demand curve forUS Treasuries is a useful indicator of market sentiment, particularly at the 30-yr maturity. A steep demand curve in 30-yr auctions is an indication that bidders are not constructive on market direction (they require higher yields/a bigger price discount to absorb incremental quantities). In these instances, we find that the scope for a sentiment shift following the auction is higher.
Conversely, a flatter demand curve at 30-yr auctions indicates a more constructive attitude in the bidding process, which tends to expose the market to the risk of a sell-off in the trading days following the auction. The slope metric has some predictive ability for bond returns, as the chart below indicates. As for 10-yr sector, the auction demand curve is important as well, although here it is the curvature, rather than the slope, that matters.
It is important to stress here that ‘price concessions’ resulting from unusually heavy supply often take place ahead of auctions. In previous work, we have shown that the effect of heavier bond supply on yields manifests itself mostly through a relative ‘cheapening’ against corresponding maturity benchmark rates, such as swaps and OIS. The market-clearing level of longer-dated yields is primarily a function of macro factors and risk sentiment. Last month, we illustrated this point through the experience of Japan during 2002-03. At the time, swap spreads went negative, on the back of higher supply, and bond yields rallied as inflation declined.
Characterising the Auction Demand Curve
We define the auction demand curve as a mapping between yields and the cumulative quantity demanded (bid) up to this yield. Put differently, the demand curve is a set of all possible pairs of yields (or prices) and demanded quantities—pretty much like any other demand curve.
In practice, Treasury auction results reveal three points on this curve—the low yield, median yield and high yield, corresponding to the first 5% of the auction, half of the auction and the full auction size. The diagram below illustrates this concept.
Using these three pairs, we can define the slope of the demand curve as:
Slope = (yield100%-yield5%)/(tender100%-tender5%)
In the same way, we can compute the upper and the lower slopes of the curve, which can then be used to
calculate the curvature:
Curvature= [Lower slope]/[Upper slope]
Two observations arise from this basic characterisation:
- Auction demand curves tend to be steeper at higher maturities. This pattern, however, disappears when adjusting for the level of yields (see the top left chart).
- The curvature of the demand curve tends to be greater at shorter maturities (with the exception of the 3-yr tenor). In other words, the steepness of the curve tends to diminish faster for lower maturities (see the top right chart).
Curves Help Predict Future Yield Movements
The slope and curvature of the demand curve contain interesting information, particularly for longer-dated maturities. Our focus is on the 10-yr and 30-yr maturities.
In order to assess the predictive ability of demand characteristics during the days following auctions, it is important to control for the macro impact on yields during this period. To this end, we focus on the residual from a linear regression of 10-yr and 30-yr maturity bonds on the 2-yr T-Note (and a constant). This residual can be loosely considered as the ‘macro-free’ component in the 10-yr and 30-yr tenors.
We now look at the predictive ability of the curve to yields after the auction. We also compare it to the predictive power of the standard auction statistics.
The one important caveat here is that the size of the sample is comparatively small as regular monthly 30-yr auctions only started in May 2009 (12 observations). The results are summarised as follows:
- A flatter 30-yr auction demand curve is typically associated with an underperformance of the 30-yr sector at all horizons considered. Indeed, the correlation between the slope and yield performance after the auction is fairly high—0.41, 0.59 and 0.32 (see the table below). The predictive ability of the slope in 10-yr auctions is less significant.
- The curvature is uncorrelated with yield changes at the 30-yr sector, but very much so at 10-yr sector. A closer examination shows that the curve is typically very concave when the low yield (the yield that fills the first 5% of the auction) is unusually depressed. This typically happens when clients ask dealers to guarantee purchases in the auction. In response, dealers submit very low bids. Indeed, ‘guaranteed delivery’ is more common at the 10-yr sector than at the 30-yr one.
- The ‘high-yield-vs-1pm’ is also found to be predictive for the 10-yr sector. The higher the high yield is relative to the pre-auction (1pm) yield (i.e., the auction results in a ‘tail’), the higher the probability that the 10-yr yield will rally in the following days (again, controlling for macro-related movements).
- The ‘bid-to-cover ratio’ is found to be irrelevant for post-auction yield performance, both for the 10-yr and for the 30-yr tenors.
- Lastly, the table above provides the historical correlation between curve statistics and the standard auction statistics. While the slope of the demand curve does have common information with the auction parameters that are widely watched, the correlation with these parameters is far from perfect. This suggests that curve statistics provide additional
information relative to those already embedded in standard auction parameters.
Shaping the Demand Curve
We find that for the 10-yr and 30-yr maturities the auction demand curve contains useful information about yield movements in the days following the auction. Moreover, other standard auction statistics, including the high-yield-vs-1-pm and the bid-to-cover ratio, do not correlate well with the demand curve, suggesting that the latter contains additional information beyond what is usually monitored by market participants.
Further investigation into what else might be extracted from these demand curve characteristics would be useful for understanding the auction dynamics. It will also be interesting to watch how the correlations evolve over time as more observations are added.