Which performs better under trader settings, double auction or uniform price auction?

A marketable permit system (MPS) has been suggested as a solution to environmental problems. Although the properties of MPSs under non-trader settings, in which each player is exclusively either a seller or a buyer, are well documented, little research has explored how MPSs perform under trader settings, in which each player can be both a seller and a buyer. We institute two auctions of trader settings in MPS experiments: a double auction (DA) and a uniform price auction (UPA). We then evaluate and compare their performances both with each other and with those under non-trader settings. The main results are as follows: DAs under trader settings perform much worse than do DAs under non-trader settings, whereas UPAs perform well, regardless of the trader and non-trader settings. UPAs are more efficient and generate more stable prices than do DAs under trader settings, and a considerable proportion of trades in DAs under trader settings consist of “flips” that could be considered speculation or errors. Thus, UPAs are likely to work better than DAs under trader settings.

1. The trader or non-trader setting is determined by whether each agent in a permit market can be both a seller and a buyer during trading periods or whether each agent can be only one or the other. If the agents can take on both roles, we denote the environment as a “trader setting,” otherwise, the environment is called a “non-trader setting” (see Ledyard and Szakaly-Moore 1994).

2. For instance, Anderson and Sutinen (2005) clearly demonstrate an event of “bubbles” in the laboratory experiments of tradable fishing allowances with DAs under trader settings.

3. Anderson and Sutinen (2006) implement laboratory experiments of tradable fishing allowance markets employing UPA and DA. However, their UPA is a “continuous uniform double auction,” which is fundamentally different from the UPA we use in our experiment. Our UPA is a sealed-bid uniform price auction, which is the same as that adopted by Smith et al. (1982) and Cason and Plott (1996), except we use trader settings. In addition, Anderson and Sutinen (2006) focus on the price discovery of fishing allowance markets and thus use different experimental parameters for UPAs and DAs. A direct comparison between the auctions cannot be made on the same ground, which is also noted by the authors. In finance, Boening et al. (1993) compare the price dynamics under DA and UPA, showing that a UPA institution reduces price bubbles.

4. The UPA under non-trader settings in our experiments follows the call market or uniform price auction introduced by Davis and Holt (1992), in which bids to buy from buyers and offers to sell from sellers are first collected and all trades are effected at a uniform price. Such UPAs have been employed for real-world trades such as in financial markets in the European, New York, American and Tokyo stock exchanges.

5. This feature is adopted to avoid “end effects,” following Cason and Gangadharan (2006).

6. For instance, a subject of \(T1\) firm is asked to submit 8 bids to buy and 2 offers to sell. Since two subjects are assigned to one type of firm among 4 types, as shown in Table 1, in total, 32 offers to sell and 48 bids to buy are submitted from 8 subjects in a period.

7. Fixed revenue was included in the payoff calculation for adjustment purposes.

8. Note that there has been no research that employs UPAs under trader settings for marketable permit experiments.

9. There are possible 6 pairs of treatments: (“\(\text {T}\_\text {DA}\) versus \(\text {T}\_\text {UPA}\),” “\(\text {T}\_\text {DA}\) versus \(\text {NT}\_\text {DA}\),” “\(\text {T}\_\text {DA}\) versus \(\text {NT}\_\text {UPA}\),” “\(\text {T}\_\text {UPA}\) versus \(\text {NT}\_\text {DA}\),” “\(\text {T}\_\text {UPA}\) versus \(\text {NT}\_\text {UPA}\)” and “\(\text {NT}\_\text {DA}\) versus \(\text {NT}\_\text {UPA}\)”). For each pair, we have run a Mann-Whitney test.

10. The Mann-Whitney test for the pair of \(\text {NT}\_\text {UPA}\) versus \(\text {T}\_\text {UPA}\) does not exhibit any statistical significance, implying that non-trader or trader settings do not affect the efficiency of UPAs.

11. In \(\text {T}\_\text {UPA}\) and \(\text {NT}\_\text {UPA}\), the average trading price and the last trading price in a period are identical.

12. There is one outlying observation of 166 at period 1 in \(\text {NT}\_\text {DA}\) for Figs. 3a and 5a. This is due to the fact that a trading price of 866 is observed at period 1 in session 4 of \(\text {NT}\_\text {DA}\) and the trade is considered an error (For the reference, all the trading prices in session 4 of \(\text {NT}\_\text {DA}\) including the outlier are shown in Fig. 8 at the “Appendix”). In Figs. 3a and 5a, we do not include the outlier to ensure the clearer visibility.

13. As was done with efficiency data in the previous subsection, we have six sessions per treatment. Therefore, we have six independent observations of average prices and the average last trading prices over 10 periods per treatment.

14. We could separately show the same figure by each period from 1 to 10. However, what we can illustrate in Fig. 6a, b is the same as the figure per period in DAs. Therefore, we decided to show only the two figures. For readers to see how trading patterns occur, especially for DAs, all the prices over 10 periods in some sessions of \(\text {NT}\_\text {DA}\) and \(\text {T}\_\text {DA}\) are shown in Fig. 9 at the appendix.

15. For robustness check, we run the regression model of Eq. (1) using the average trading prices per session in each period. The results are almost the same as that shown in Table 4. The estimated convergent average price in \(\text {T}\_\text {DA}\) is 95, which is outside the equilibrium price range. The estimated convergent average prices in other treatments are within the equilibrium price range or sufficiently close to it.

16. Godby (2002) conducts various treatments of market power experiments for emission permit trading. Among them, treatment 5 is the closest to the experiments in our research, generating a TR of 1.9. Because the TR in our experiments is 1.8, the TR result seems to be plausible and consistent with Godby (2002).

17. In asset market experiments, Lei et al. (2001) demonstrate that the occurrences of bubbles and poor performances in DAs are not due to speculative trades. There are two differences between our results and those of Lei et al. (2001). One difference is that each permit in our experiment is a single period-lived asset and possesses a clear underlying value of MACs, whereas a commodity in the asset market is a multiple period-lived asset, and its underlying value depends on subjects’ future expectations of dividends or returns. Therefore, the underlying value of multiple period-lived assets changes in a more complex manner than that of permits in the MPS experiments. The other difference is that the non-trader settings in this research can be considered to give information about which side (buyers or sellers) of markets each subject is on, whereas the “no-resale” treatment in Lei et al. (2001) does not give such information.

18. Jamison and Plott (1997) demonstrate that bids and asks in the closing moments of a continuous double auction play a fundamental role in price discovery. Easley and Ledyard (1993) develop a theoretical model of double auctions by assuming that the last trading price for a period in a continuous time double auction contains key information for price adjustment in the next period.

The authors thank anonymous referees, Makoto Kakinaka and Hiroaki Miyamoto for their helpful comments, advice and supports. We are also grateful to the financial supports from the Japan Society for the Promotion of Science (JSPS) as the Grant-in-Aid for Scientific Research B (16H03621), JSPS Specially Promoted Research and Ministry of Environment, Japan (S-16) and Kochi University of Technology.

Authors and Affiliations

1. Research Institute for Future Design, Kochi University of Technology, Kochi-shi, Japan

   Koji Kotani

2. School of Economics and Management, Kochi University of Technology, Kochi-shi, Japan

   Koji Kotani

3. Urban Institute, Kyusyu University, Fukuoka, Japan

   Koji Kotani

4. College of Business, Rikkyo University, Tokyo, Japan

   Koji Kotani

5. Faculty of Economics, Musashi University, Tokyo, Japan

   Kenta Tanaka

6. Departments of Urban and Environmental Engineering, School of Engineering, Kyushu University, Fukuoka, Japan

   Shunsuke Managi

Corresponding author

Correspondence to Koji Kotani.

Appendix

In this appendix, all the trading prices over 10 periods in some sessions of \(\text {T}\_\text {DA}\) and \(\text {NT}\_\text {DA}\) are shown in Figs. 8 and 9 as samples for readers to see how trading patterns occur, especially for DAs.

All the trading prices over 10 periods in session 4 of \(\text {NT}\_\text {DA}\)

Full size image

Samples of all the trading prices over 10 periods in some sessions of \(\text {NT}\_\text {DA}\) and \(\text {T}\_\text {DA}\). a All the trading prices over 10 periods in session 4 of \(\text {NT}\_\text {DA}\), b all the trading prices over 10 periods in session 1 of \(\text {T}\_\text {DA}\)

Full size image

Reprints and permissions