Standard Deviation of Bids for Construction Contract Auctions

Abstract

Previous research has confirmed that the distribution of bids for construction auctions can be reasonably modelled with the Lognormal distribution. The location parameter of this distribution (the mean μ) has been found to have a good linear correlation with the bidders’ cost estimates. However, the scale parameter (standard deviation of the bids, σ) remains noticeably difficult to anticipate. By analyzing 13 construction auction datasets, hard evidence is provided that the high variability of σ observed in construction auctions is mostly due to sample size (number of bids per auction). Moreover, we show that the coefficient of variation (σ/μ) of log-transformed bids follows the same χ² distribution in uncapped auctions. This means the σ's population value in similar auctions is nearly proportional to μ provided the bid price is not upper limited. Other findings are that more frequent bidders do not tend to bid lower, but their dispersion is narrower than sporadic bidders. These findings allow the introduction of important simplifications in construction bidding models, especially when access to historical data is limited.