Central bank liquidity and “toxic asset” auctions

Paul Klemperer 25 September 2009



The crisis began in early August 2007, and a bank run led to Northern Rock’s collapse in mid-September. The Bank of England wanted urgently to supply liquidity to banks and was therefore willing to accept a wider-than-usual range of collateral, but it wanted a correspondingly higher interest rate against any weaker collateral it took.

A similar problem faced the US Treasury during its autumn 2008 Troubled Asset Relief Program where it planned to spend up to $700 billion on “toxic assets” with a face value well in excess of $1 trillion (Pagano 2008). There were on the order of 25,000 closely related but distinct securities and perhaps 300 likely sellers, but the largest ten sellers held something like two-thirds of the toxic assets.

Complicating matters in both cases, the pace of financial markets required any auction to take place at a single instant, thus ruling out many of the multistage auction techniques used in other areas. The problem with multi-stage auctions in the financial world is that bidders who entered the highest bids in early stages might change their minds about wanting to be winners before the auction closed. Moreover, financial markets themselves might be influenced by the evolution of the auction creating opportunities for manipulation.

How to proceed: A new auction design

How should goods that both seller(s) and buyers view as imperfect substitutes be sold, especially when multi-round auctions are impractical? This column outlines a new solution to all these problems – the “Product-Mix Auction” (so-called because it solves the general problem of a firm that can offer multiple product varieties to customers with different preferences, subject to capacity and other constraints).

My design is straightforward in concept – each bidder can make one or more bids, and each bid contains a set of mutually exclusive offers. Each offer specifies a price (or, in the Bank of England’s auction, an interest rate) for a quantity of a specific "variety". The auctioneer looks at all the bids and then selects a price for each "variety". From each set of offers in each bid, the auctioneer accepts the one that gives the bidder the greatest surplus evaluated at the selected prices or no offer if all the offers would give the bidder negative surplus. All accepted offers for a variety pay the same (uniform) price for that variety.

The idea is that the menu of mutually exclusive bids allows each bidder to approximate a demand function, so bidders can, in effect, decide how much of each variety to buy after seeing the prices chosen. Meanwhile the auctioneer can look at demand before choosing the prices. (Allowing the auctioneer to choose the prices ex post creates no problem here because it allocates to each bidder precisely what that bidder would have chosen given those prices in the environments for which the auction is proposed.) Importantly, offers for each variety provide a competitive discipline on the offers for the other varieties, because they are all being auctioned simultaneously.

Comparing a product-mix auction with existing approaches

Compare this with the three "standard" approaches:

The first traditional approach is to run a separate auction for each different “variety”. In this case, outcomes are erratic and inefficient, because the auctioneer has to choose how much of each variety to offer before learning bidders' preferences, while bidders have to guess how much to bid in each auction without knowing what the price differences between varieties will turn out to be. The wrong bidders may win, and those who do win may be inefficiently allocated across varieties.

Furthermore, each individual auction is much more sensitive to market power, manipulation, and informational asymmetries than if all offers compete directly with each other in a single auction. The auctioneer’s revenues are correspondingly generally lower. Thus, for example, if the US Treasury had simply predetermined the amount of each type of security to purchase, ignoring the information about demand for the large number of closely related securities, competition would have been inadequate because of the highly concentrated ownership of the assets. All these problems also reduce the auctions’ value as a source of information. They may also reduce participation, which can create "second-round" feedback effects further magnifying problems.

A second common approach is to set fixed price supplements for “superior” varieties and then auction all units as if they are otherwise homogeneous. This can sometimes work well, but such an auction cannot take any account of the auctioneer’s preferences about the proportions of different varieties transacted. In any case, a central bank might not want to disclose its view of appropriate price-differentials for different collaterals to the market in advance of the auction. Furthermore, the auctioneer suffers from adverse selection. If, for example, the US Treasury had simply developed a "reference price" for each asset, the bidders would have sold it large quantities of the assets whose reference prices were set too high – and mistakes would have been inevitable, since the government had much less information than the sellers.

The final approach that is sometimes used is the simultaneous multiple round auction – the multi-stage auction in which bidders take turns bidding on multiple assets until no one wants to bid again on any asset. My product-mix auction yields similar outcomes but is more robust against collusion and other abuses of market-power. Furthermore a simultaneous multiple round auction is often infeasible – especially in financial markets – because of transaction costs, the time required to run it, or because its complexity is too off-putting to bidders.

In Klemperer (2009), I show how my new approach – the product-mix auction – can be implemented, and that it is simple, robust, and easy for bidders to understand, so that they are happy to participate.

The product-mix auction yields better “matching” between suppliers and demanders, reduced market power, greater volume and liquidity, and therefore also improved efficiency, revenue, and quality of information than feasible alternatives. Its potential applications therefore extend well beyond the financial context.


Klemperer, Paul (2009). “The Product-Mix Auction: A New Auction Design for Differentiated Goods”.

Pagano, Marco (2008). “What is a reverse auction?”, VoxEU.org, 21 October.



Topics:  Financial markets Global crisis

Tags:  auctions, toxic assets, global crisis

Edgeworth Professor of Economics at Oxford University and CEPR Research Fellow