«CHINA'S LAND MARKET AUCTIONS: EVIDENCE OF CORRUPTION Hongbin Cai J. Vernon Henderson Qinghua Zhang Working Paper 15067 ...»
As with English auctions, two stage auctions are announced about 20 working days in advance; details of the plot are made public; and a deposit is required upon participation in the auction. A key difference is the auction format. With this type of auction, there are two stages. The first stage normally lasts 10 working days after the auction starts. In the second stage, at the end of the 10 working days, if more than one bidder is competing for the property, the auction ends on the spot with an English auction where only active bidders in the first stage are allowed to participate. If there is an ending English auction, it is generally less public than regular English auctions. In the first stage The third type is sealed bid, or zhaobiao auction. There, bidders submit sealed bids to the land bureau, which decides the winner according to a complicated score function, in which the submitted bid usually accounts for only 20-30% in weight. The remaining 70-80% of the weight goes to the credibility of the bidder and how much social responsibility the bidder is willing to take. Credibility is mainly reflected in two aspects: one is the quality and reputation of the projects the bidder has developed in the past; the other is the bidder’s financial capacity. As for social responsibility, this arises from Beijing’s recent attempt to curb rising housing prices.
during the 10 working days between the starting date of the auction and the potential ending English auction, after obtaining qualification, people may submit ascending bids in person or on-line. Bids as they arrive are immediately posted on the trading board of the land bureau, as well as typically on the internet, although the identity of bidders is not posted. If, at the end of 10 working days, there is only one remaining bidder, that bidder is assigned the property at his bid price (but not less than the reserve price). Otherwise, with competition, the auction is converted to an English auction.
We will argue this first stage of a two stage auction allows for signaling of valuations for purposes of entry deterrence in non-corrupt contexts and for signaling in a potentially corrupt context, where the auction will be dominated by a corrupt bidder (in league with the land bureau). 4 There are two fuzzy parts to the two stage auction format which we will argue are consistent with signaling, especially under corruption. While the auction is announced about 20 working days in advance, the exact date of the start of the first stage of the auction may not be announced at that time, but rather at an unspecified later date. Second, while bidders can apply during the announcement period before the first stage starts, it appears that approvals to participate can be delayed until after the first stage is under way.
As detailed below, we use data on 2302 auction transactions from 2003 to 2007 in 15 cities, which use both auction types (as opposed to having only two stage auctions). In these cities English auctions account for 28% of auctions. In Figures 1 and 2 we present the indications from raw data that properties sell at a higher price under English as opposed to two stage auctions, and that English auctions are much more likely to be competitive. Figure 1 shows the distribution of unit sales price (price per sq. meter), by auction type for the sample; and Figure 2 shows the distribution of the ratio of sales to reserve prices. The raw data suggest that the distribution of unit sales prices for English auctions is shifted to the right of that for two stage auctions. Of course, the distribution of raw unit sales prices does not condition on property and market characteristics. While We conducted surveys of the land bureau officials of 20 cities covered in our sample. In our survey, we asked questions regarding the differences in the mechanism between the two auction formats, in addition to differences in the pattern of bidding behaviors. We also asked how the land bureau chooses between the two auction formats for each piece of land for sale. Our theory conjectures in the next section are informed by our survey findings.
differentials in property characteristics could explain price differentials across auction types, below we show that will turn out not to be the case.
In Figure 2, two stage auctions tend to be massed much more around 1.0 for “spread”, which is the ratio of sales to reserve prices. Both conceptually and in a particular sub-sample analyzed below where we know the number of bidders, a ratio of 1 implies that there is just one bidder and thus no competition. Ratios larger than 1 in the sub-sample imply multiple bidders and what we term a competitive auction. Of course, whether there is competition or not is influenced by reserve price; so, if there are differentials in setting reserve prices across auction types, that might explain the differential pattern by auction type in Figure 2. However, reserve prices are set by the outside committee before the choice of auction type by the land bureau; and, as we will see below in Table 4, reserve prices do not seem to affect the choice of auction type. 5 Assuming that reserve prices are some fraction of an assessor’s estimate of true market value, a lack of competition is very surprising on its own. In these cities, auctions occur in a setting of rapid urban growth, with per capita urban incomes growing at about 10% a year and local population at 3-4 % a year. Given national restrictions on conversion of rural to urban land at the city fringes, this suggests there should be a high demand for land for new development.
In the next section we outline a simple conceptual framework, which discusses our empirical hypotheses and the framework behind them. In the following sections, we turn to the econometric formulation looking at auction choices, sales prices, and the degree of competition. We document and explore the price differences between English and two stage auctions. As already noted a key issue concerns selection of properties into auction type and whether properties with better unobservables are more (positive selection) or less (negative selection) likely to sold by two stage auctions.
As further evidence that reserve prices are set independently of auction type, in a MLE Heckman selection model where we allow auction type to be a determinant of reserve price, controlling for selection effects of auction type yields an insignificant rho, suggesting no correlation between those unobservables affecting auction choice and those affecting reserve price.
2. Conceptual framework We have not fully solved a theoretical model which covers all key aspects, for reasons that will become apparent. Rather we discuss the elements that are relevant and try to draw reasonable conjectures. We start with a discussion of how the two auction formats might be modeled in a world without corruption and how the choice of format affects outcomes. Then we ask how corruption possibilities might have differential implications for the two formats. Finally we discuss the land bureau’s choice of auction format.
Assume for a leasehold auction there are N potential bidders, of which some endogenous number n pay an entry fee, C, and become active bidders. 6 A key issue is how the choice of auction format may influence n. We assume auctions are independent private valuation. Specifically, a potential bidder i ’s valuation is Vi = v0 + vi, where v0 is the (expected) common value that is the same for every bidder (based on property characteristics and local market conditions) and vi is the private value component only known to bidder i. vi ’s are i.i.d. 7 We make the standard assumption that all bidders are risk neutral and maximize their expected payoff. Let Vi ∝ F(V) on [v0, V ] be the distribution function of the bidder i’s valuation, and let f(V) be the associated density function. A bidder’s payoff when winning the auction with a bid Bi is U i = Vi − Bi − C.
2.1 English auction.
Since an English auction is outcome-equivalent to a second price Vickery auction, our setting is equivalent to that of Tan and Yilankaya (2006), who analyze a simultaneous move entry game in a second price auction with independent price valuations and participation costs. In a symmetric equilibrium of such a model a bidder will decide to ˆ enter the auction if and only if his valuation is above a certain value V r + C, where r is the reserve price and C is the entry cost. For a bidder with valuation exactly equal to ˆ V, the only way he can get positive rent from entering the auction is if he is the lone The entry fee consists of (i) cost of making cash deposit to qualify, (ii) cost of preparing documents to meet the qualification requirements, (iii) other transaction costs (e.g., time, consulting fee).
Note that we assume the (expected) common value is common knowledge to all participants, thus the auction is treated as one with independent private valuation.
possible outcomes in the auction are (1) that there are no bidders, which occurs with probability F (V ) N ; and (2) that there are two or more bidders, so the auction is ˆ
potentially signal that she drew a relatively high private valuation; we will suggest that the bid will signal her actual valuation. The point of the signal is to discourage future potential entrants who might have drawn somewhat higher valuations from entering the auction. For future bidders with somewhat higher valuations, they now know that, if they enter, the prior signaler is prepared to bid up to her valuation. That inferred valuation then defines the minimum price they have to pay. So their expected rent is based on the likelihood that they will win (no future entrants will have an even higher bid) and on the ex post rent (their valuation minus the current signaler’s valuation). Solving the general case with endogenous first stage entry sequence is daunting—whether an early entrant signals with what bid function, whether later entrants with higher valuations enter or not, and the complicated interactions between early and later signalers.
We illustrate the setting with a simple example. Suppose only one randomly selected person has the option to bid early and we consider a possible separating signaling equilibrium. Suppose bidder 1 chooses to enter in stage 1 by using a strictly increasing bidding schedule B(V1 ) when her valuation is V1 ∈ [V, V ]. For V1 V, bidder 1 will choose to not enter the auction.
Suppose her valuation is exactly V. Based on the Riley argument in the signaling literature, bidder 1 should use the lowest possible signal, the reserve price r. Moreover, in order for bidder 1 not to choose this signal when her valuation is just below V, her expected payoff from entering and signaling must be zero. Once bidder 1 reveals that her ˆ valuation is V, other potential bidders will enter only if their valuation is above VS (V), the solution to equation (1) with V replacing r and N − 2 replacing N − 1. Other potential bidders understand that to win they must outbid bidder 1; but since bidder 1 has committed to enter the auction, she is willing to bid all the way to V1. Thus, for the other potential bidders, the effective reserve price increases to V1. Bidder 1 can win the auction only if no other potential bidders enter the auction, which happens with probability of F(VS (V))N −1. Therefore, we must have ˆ
ˆ ˆ VNS V(r, C, N,V), which is the equilibrium entry threshold in the case of an English auction with N potential bidders.
In the Appendix, we show that if bidder 1 has valuation V1 ∈ [V, V ], he has a bidding function that is strictly increasing in V1 and truthfully reveals his valuation. Such a bidding function satisfies the single crossing property, so it isn’t beneficial for lower valuation bidders to pretend to be higher types.