How should government bonds be sold? Research typically emphasizes how the auction design affects outcomes depending on the nature of demand and the competitive environment. This study combines models of strategic bidding in Treasury auctions with detailed bidding data to construct empirical measures that reveal the effectiveness of auctions. Applying these methods to data on US Treasury auctions shows that the gains from optimizing the auction mechanism are no more than 5 basis points. The research also quantifies the advantage enjoyed by primary dealers in these markets, who are able to observe the ‘willingness-to-pay’ of their customers who route their bids through them.
When governments need to raise money to finance their operations – either to synchronize tax receipts with expenditures or to cover deficits – they turn to financial markets. They typically issue securities of various maturities: short-term Treasury bills, with maturity up to a year; medium-term notes, with maturity up to ten years; and long-term bonds, with maturity up to 30 years. The goal of such security issuances is to raise as much revenue as possible, which essentially means to sell a given amount of securities at the lowest ‘discount’ (or yield) possible.
Organizing the bond markets
There are two important considerations when deciding how to organize the markets for these securities:
- First, how to run the initial offerings: should an auction be used – and if so, what type?
- Second, whether or not to restrict access to the initial offerings – and if so, how to regulate access?
While there is a general consensus among economists that government securities should be sold via auctions, there is little agreement on what type of auction to employ. Bartolini and Cottarelli (1997) report that 39 out of 42 countries use the ‘pay your own bid’ or discriminatory auction, and the remaining three use a ‘single price’ or uniform price auction.
There is little agreement on what type of auction to employ
In both auction formats, the auctioneer allocates the issuance to bidders according to a descending order of bids. In a ‘pay your own bid’ auction, each bidder has to pay the full amount of their bid for any security that they are allocated. In a uniform price auction, all bidders pay the same price, equal to the lowest satisfied bid.
Many economists suggest that the uniform price auction should encourage participation, especially for small bidders who do not have to be worried about paying too high a price relative to other participants if they bid their true ‘willingness-to-pay’. Since everyone pays the same (market clearing) price, that is the most they would pay.
But large bidders in uniform price auctions have an incentive to exercise market power to lower the market clearing price. If they act strategically and reduce the size of their order, they may pay a lower market clearing price for the units they purchase.
On the other hand, in a discriminatory auction every bid that is filled will have to be paid up. Strategic bidders must guess where the market is likely to clear. If they lower their bid, they pay less when the bid is filled, but a lower bid will not be accepted if it falls below the market clearing price.
Unfortunately, except in special environments, economic theory does not make unambiguous predictions for which auction format leads to higher revenues for the auctioneer. We therefore turn to data to address the question of auction format choice.
There is a burgeoning body of research evaluating auctions of government securities in many countries: Austria, Canada, the Czech Republic, Korea, Mexico, Turkey, the United States and others. Early work studied countries that switched from one mechanism to another like the United States (Nyborg and Sundaresan, 1996) or Mexico (Umlauf, 1993), but the comparison before and after the switch is valid only under restrictive assumptions about the stability of the economic environment.
Our strategy, developed in our previous work, combines economic theory and econometrics: we first use microeconomic theory and econometric methods to estimate the willingness-to-pay of bidders in these auctions. We then use theory to tell us what the outcomes would have been under a different auction mechanism.
Below we discuss the intuition behind our approach, which we developed in our prior work and which combines a model of bidding in these auctions with the data in order to evaluate whether or not the issuer could do better by changing the mechanism.
As to who should be allowed to bid in these auctions, historically, most markets for government debt have been organized around a group of large and important players, known as primary dealers. These players, which are typically large financial institutions or investment banks, are charged with making the market for government securities. They are required to quote bid-ask spreads for the securities so as to provide liquidity and to participate actively in the initial security offerings by submitting bids.
In exchange for this, they are given exclusive access to the primary market – the initial security issuance. Hence, anyone wishing to purchase securities during the initial issuance has to route their orders through one (or more) of the primary dealers.
Since the late 1990s, the United States Treasury allows anyone to participate in the auctions directly (after fulfilling registration and collateral requirements). Nevertheless, the US Treasury still relies heavily on the primary dealer system: many features of the system and any potential changes are discussed with them prior to implementation.
We use the data to perform a back-of-the-envelope calculation of how much primary dealers gain on average from having access to the bids of their customers. Some of the customers can be important players in the financial markets themselves – the investment management firm BlackRock, for example. Hence, having access to their orders allows the primary dealers to gauge the market sentiment before submitting their own bids to the auction.
Effectiveness of allocation and revenue extraction
To conduct our analysis, obtaining data from a set of conducted auctions is not enough. As is well known to economic theorists, the uniform price auction format used by the US Treasury creates strategic incentives for bidders to understate their true willingness-to-pay.
First, we need to get a sense of how far the auction bids are from bidders’ true willingness-to-pay. In a close to perfectly competitive auction environment – one in which bidders cannot affect the market clearing price by modifying their bid – they would submit bids that are close to their actual willingness-to-pay. This ensures an efficient allocation, in which bidders who value the securities the most are the first to be allocated.
But since in practical settings, the sets of competitors are limited (for example, to the set of primary dealers), sophisticated bidders may have substantial room to act strategically and ‘shade’ their true willingness-to-pay in order to maximize the surplus they gain from the auction. Whenever they start shading their bids, however, the auction is no longer guaranteed to allocate the securities to the bidders with the highest willingness-to-pay, which results in unnecessary inefficiency.
To quantify these two problems from which an auction mechanism can suffer, we combine the data with a model that captures the strategic incentives that the bidders are facing when submitting their bids. This model then allows us to associate the observed bids with the willingness-to-pay that is rationalizing them.
Because primary dealers’ demands are larger, they have the most to gain from a lower market clearing price and hence they ‘shade’ their bids mostTo address the second key question fully – whether to restrict access – one would need to evaluate the costs and benefits of the primary dealer system. The trade-off is between, on the one hand, higher yields than necessary due to restricting competition, which raises the cost of debt issuance, while, on the other hand, guarantees of a minimal level of liquidity provided at any point in time.
There is certainly room for more work in this area – for example, to quantify the effect that leaving the primary dealer system and instead allowing free participation would have on liquidity. Our work studies a more modest question: how much do the primary dealers gain from smaller players (we can call them customers) being required to route their bids through them?
We use data from uniform price auctions of US Treasury bills and notes between July 2009 and October 2013. The first fact to note is that primary dealers consistently bid higher yields (lower prices) than direct or indirect bidders.
But this does not immediately mean that primary dealers value the securities less than other categories of bidders: it is possible that primary dealers, due to their larger size and market share in the auction, are able to exercise more market power, and therefore shade their willingness-to-pay more than other, smaller bidders.
Using our data, we estimate a structural model of bidding to recover the marginal willingness-to-pay that is underlying participants’ bids. In doing so, we respect the institutional features of the bidding system: that primary dealers observe bids submitted by their customers.
We find that on average, primary dealers have higher willingness-to-pay than direct and indirect bidders; this is in stark contrast to what is suggested by naïve analysis of bidding patterns, which may suggest that primary dealers have lower willingness-to-pay. Because primary dealers’ demands are relatively larger, they also have the most to gain from moving the market clearing price and hence they shade their bids most.
Putting the numbers in context
The numbers we report are not uncommon in studies of Treasury auctions. Our estimates of willingness-to-pay also enable us to quantify the amount of efficiency loss arising from misallocation of securities due to strategic behavior.
We find that efficiency losses are modest, less than 1 basis point for short-term securities to 6.4 basis points for 10 year notes, averaging to just above 2 basis points. This implies that improvements in allocative efficiency due to a change in the auction mechanism are going to have an upper bound of 2 basis points on average.
Our results also allow us to put an upper bound on the amount of revenue that the US Treasury can hope to gain from changing the format of its auctions. Since the revenue gain from any auction mechanism should be below the revenue gain from switching to a (hypothetical) mechanism that captures the entire surplus, the amount of revenue gain from switching auction format should be less than the total of bidder surpluses and efficiency losses.
This suggests that the maximum revenue gain from switching auction format is 5 basis points. While 5 basis points corresponds to $350 million over a total auction volume of $700 billion per year, we should note that this is a very conservative upper bound: capturing all bidder surplus is never achievable due to market participants playing strategically, no matter what mechanism we employ. (Moreover, if such a mechanism were to be employed, bidders would not participate in the auction.)
The maximum revenue gain from the US Treasury changing its auction format would be 5 basis points
This is simply a consequence of bidders having private information: bidders know in a given week how much they are willing to pay for a given security, and the issuer has to award them some surplus to reveal this information in the auction. In the applications cited above, similar magnitudes have been reported whether a country uses a discriminatory auction or a uniform price auction.
Perhaps more importantly, the danger from upsetting a market that is accustomed to a certain way of operating might be more dangerous and potentially costlier to a smooth funding of government expenditures than a little bit of surplus left to the participants.
A few caveats
It is important to keep in mind that the model through whose lens the data are interpreted relies on several assumptions. While there are no functional form restrictions on bidders’ payoffs, there is an important assumption about information structure.
In the language of auction research, one of the maintained assumptions is that bidders have private values. This essentially implies that if dealers were to learn the private information that rivals possess about their willingness-to-pay for the security, it would not affect their own willingness-to-pay.
This assumption is fine in situations where all information on the value of the security that is relevant for everyone is aggregated into some publicly available signal prior to the auction (such as a forward price or when-issued price, which is an active market in the United States prior to the auction). It is important to note that even when values are private, primary dealers would benefit from learning about demands (bids) of other auction participants, because they would be in a better position to guess the price at which the market will clear.
In a discriminatory auction, this would result in a lower payment since there would be no need to submit high bids to guarantee winning. In a uniform price auction, primary dealers would have a better idea how to tailor their bid to make the market clearing price even lower – hence lowering the payment for everyone.
Another potential issue that our work does not address is collusion. Recently, there have been several cases of market manipulation (LIBOR, FX), which involved financial institutions that are also primary dealers for US government securities.
This article summarizes ‘Bid Shading and Bidder Surplus in the US Treasury Auction System’ by Ali Hortaçsu, Jakub Kastl and Allen Zhang, published in 2018 in the American Economic Review 108(1): 147-69.