Testing market regulations in experimental asset markets – The case of margin purchases

Abstract Margin requirements regulate the risks of leveraged positions in financial markets. Violated margin requirements trigger margin calls leading to automated liquidation of open margin positions. Due to a lack of active margin regulation, however, empirical studies are not able to capture the effect of margin regulation on asset market performance. Instead, we conduct an experimental finance study to understand how margin regulations, and in particular margin purchases, influence market performance. Our experimental results indicate that permitting margin purchases inflates market prices; in fact, active margin trading positively correlates with market prices. In a robustness check, we also permit short sales which curb though not eliminate the effects from margin purchases.

Eventually, trading had finished when traders realised being bankrupt, which does not affect the market anymore. Instead, in our setup, the credit line is fixed exogenously but depends on the market value of shares; the higher the share price, the higher the credit line and vice versa. In our design, the inflow of additional cash is endogenously determined through margin purchases as our debt capacity depends on the value of assets, i.e. on the current market prices. As the ability to borrow money increases with collateral, the liquidity argument does not hold in our markets. However, liquidity seems to dry out rapidly after the bubble crashes as margin calls and resulting forced asset sales end in a lower value of collateral triggering even more margin calls.
Further on, we consider short selling as a robustness check for the effect of margin purchasing on market performance. In contrast to other experiments, we do not limit the size of short positions ( Haruvy and Noussair, 2006;Ackert et al., 2006;Duchêne et al., 2019 ), but allow the short sale capacity to fluctuate with market prices analogous to our margin purchasing implementation in real world markets.
Our contribution adds to the recent experimental studies on financial regulation policies. Fischbacher et al. (2013) consider monetary policy regimes in experimental asset markets. Armantier et al. (2013) experimentally tested "Reference Price Auction" mechanisms to remove toxic assets from portfolios of financially stressed banks (TARP auctions). Keser et al. (2017) implemented a game-theoretic model on the interaction between issuers and credit-rating agencies. Also, Weber et al. (2019) use the laboratory as a test-bed to analyse credit default swap regulation. In another paper, we also test different forms of IPO mechanisms .
We proceed as follows. In Section 2 , we consider our first experiment in which we consider the difference between asset market performance in markets with and without the ability to buy on margin. In Section 3 , we consider our second experiment in which we additionally allow for short selling in both settings. We conclude in Section 4 .

Experimental implementation
We make use of the asset market design of Smith et al. (1988) , which is known to produce speculative price bubbles in the laboratory ( Palan, 2013;Janssen et al., 2019 ); the appendix provides detailed instructions. 2 Nine subjects, endowed with shares and cash (see Appendix B.3 ), trade 18 shares of a fictitious company during a sequence of 15 double-auction trading periods, each lasting three minutes. At the end of every period, each share pays a dividend that is 0, 8, 28, or 60 cash units (Taler) with equal probability. Since the expected per-period dividend equals 24 Taler, the fundamental value in period t equals 24 × (16 − t ) , and declines from 360 in period 1 to 24 in period 15. After the last dividend payment in period 15, the shares become worthless and the cash holdings-the endowment plus gains from trade plus dividends-are paid out at an exchange rate of one cent to one Taler. The instructions include a table which explains the dividend process, including the sum of the expected dividend payments ( t × 24) and the range of the sum of dividend payments ({ t × 0, t × 60}).
We implemented an empirically relevant trading facility, where traders submit limit and market orders to an orderdriven continuous electronic market featuring an open order book. We thus deviate to some extent from the usual double auction implementation to allow for margin calls and also for cancellation of orders. 3 When a trader wants to sell a share, she submits an offer to sell at a minimum price. A trade occurs immediately if her submitted sell price undercuts the highest standing buy price (a market order); the price then equals the highest standing buy price, which is instantaneously removed from the order book. Otherwise, the limit order is listed in the open book of sale orders (where lower orders are listed before higher ones). 4 When a subject wants to buy a share, she submits an offer to buy for a maximum price. A trade occurs immediately if the submitted buy price exceeds the lowest standing sell price (market order); the price equals the lowest listed sell price. Otherwise, the limit order is listed in the open order book of purchase orders (in decreasing order). 5 Subjects were also able to cancel their standing limit orders.
To evaluate the impact of margin purchases on asset market prices, we compare the control treatment without margin trading opportunities ( C ) to a setting in which margin purchasing is permitted (Buy on Margin -B ). Traders execute margin purchases when they purchase shares on loan (negative cash balance), which is collateralized by the shares (collateral = shares × current market price). The debt capacity depends on the value of collateral, i.e. traders can take up debt as long as the debt to collateral ratio is below 50% ( = maintenance margin). On the one hand, a price increase enhances the 2 The chosen asset market design has been implemented in several studies ( Palan, 2013;Powell, 2016 , e.g.) and some concerns have been raised (e.g., Kirchler et al., 2012 ). However, the Smith et al. (1988) asset market design still serves as the main work horse most prominently shown in recent studies like e.g. ( Charness and Neugebauer, 2019;Carlé et al., 2019;Kocher et al., 2018;Füllbrunn et al., 2019 ). Using a different asset market design based on a flat fundamental value (dividend payments and interest balance out) introduced by Holt et al. (2017) and Weitzel et al. (2020) consider experiments with students and financial professionals and shows that market institutions comparisons like permitting short selling yield qualitatively similar results as the Smith et al. (1988) asset market design.
3 See for example http:// http://www.nasdaqtrader.com/Trader.aspx?id=Workstation . However, we do not use a market maker. Usually, asset market experiments are double auctions without the possibility to cancel the order. 4 For example, subject A submits a sell order at 220 and subject B has the highest standing buy order at 300. Then, subject A sells to subject B at a price of 300. 5 For example, subject B submits a buy order at 300 and subject A has the lowest standing sell order at 220. Then subject A sells to subject B at a price of 220. JID: JEBO [m3Gsc;October 12, 2020;21:55 ]  collateral, and thus more shares can be purchased on margin. On the other hand, a price decline devalues the collateral while the debt remains constant. As soon as the debt to collateral ratio exceeds 50% a margin call is triggered. Immediately, i) the trader's buttons are disabled, ii) outstanding orders are canceled, and iii) the computer automatically quotes a limit sell order at the prevailing market price. The computer continuous selling shares until the margin requirements hold again, or until all shares have been sold. 6 Note that in contrast to other experimental asset market designs our design enables cash in-and outflows dependent on the market price, incorporating margin calls and forced asset sales. A margin call can lead to bankruptcy. However, the consequences of a margin call hold even during bankruptcy, i.e. outstanding positions are continuously being closed although subjects are bankrupt. One caveat with our implementation is that high dividend payments reduce debt quite substantial which reduces the threat of margin calls and the risk of bankruptcy.

ARTICLE IN PRESS
In the experimental sessions, subjects interact when inexperienced (round 1), and when once-experienced (round 2) within the same cohort under identical conditions. Before the first round started, subjects participated in training periods and answered a computerized questionnaire. Payments equalled the final cash balance from both rounds exchanged to euro (100 Taler = 1 Euro) plus a show-up fee. We conducted the experiments at the MaXLab (University of Madgeburg) and the BonnEconLab (University of Bonn) using z-Tree (Fischbacher 2007) for the programming and ORSEE (Greiner 2004) for the recruitment. One hundred thirty-five students participated in seven C sessions and eight B sessions. Including comprehension questionnaires and payment, the session lasted up to 3.5 hours. The sessions paid on average 30 euro in C (SD 8.74) and 31 euro in B (SD 9.84) including show-up fee.
While textbook theory neglects loan defaults, recent models with heterogeneous agents argue that 'natural buyers' (or optimists) inflate prices when having sufficient funds from borrowing money ( Fostel and Geanakoplos, 2008 ). Permitting margin purchases endows the natural buyers with sufficient liquidity to push up prices. Limiting the access to additional funds dampens inflationary pressures though. Thus, our primary hypothesis is that asset prices are higher under treatment B in which buy on margin is permitted than under the baseline treatment C .

Experimental results
Fig. 1 depicts the deviations of median prices from fundamental value for the treatments B and C by period and round. The (red) straight line represents the fundamental value, and positive values indicate prices exceeding the fundamental value. The grey curves show the trajectory for each session separately, while the black curves with diamonds show the treatment average for each period. In the control treatment, the trajectories form the standard pattern observed in these experiments, i.e. prices start below fundamental value, increase and intersect the fundamental value, reach the peak, and finally get back to fundamental value. The pattern is less pronounced in round 2, when subjects had more experiences. With margin purchases, the pattern is somewhat different as prices do not start below fundamental value but at or above fundamental value; a reason might be the higher liquidity in comparison to C . In round 2, we can see that prices are still above fundamental value, but only in the early periods; in later periods, prices follow fundamental value in B and also in C . In line with Haruvy et al. (2007) the price peaks move to earlier periods in round 2. Already from the start, we can see that prices are higher in B than in C ARTICLE IN PRESS JID: JEBO [m3Gsc;October 12, 2020;21:55 ]  To measure the deviation from fundamental value, we consider Relative Deviation (RD) and Relative Absolute Deviation (RAD) as suggested by Stöckl et al. (2010) , Positive Deviation (PD) as suggested by Eckel and Füllbrunn (2015) , and the highest deviation from fundamental value (PMAX). 7 When the relative deviation is 10%, then we say that the market is overpriced by 10 percent. When the relative absolute deviation is 15%, we say that the market is mispriced by 15%. Positive deviation equals the area between fundamental value and prices when prices are above fundamental value. Hence, it measures the absolute magnitude of overpricing in Taler. In the appendix section Appendix A.1 , we define the applied measures. Additionally, we also consider the highest amount borrowed in a market (CREDIT). Table 1 reports the average outcomes of the above defined measures for each round and also the differences between rounds.
Overpricing in B is about 70% in round 1 and still 33% in round 2-significantly positive in both rounds ( p 1 = 0 . 012 , p 2 = 0 . 0357 , using a sign rank test) but also significantly lower in round 2 ( p = 0 . 050 ). Overpricing in the control C is much lower: 22% in round 1 and merely 5% in round 2-significantly lower in round 2 ( p = 0 . 028 ). While overpricing is marginally significant in round 1 ( p = 0 . 063 ), it is no longer significant in round 2 ( p = 0 . 500 ). RAD shows a similar pattern as does PD; in both cases, the measured values decline with experience. However, the difference in peak prices (PMAX) is not significant between repetitions. Overall, we can conclude that the magnitude of overpricing is reduced from round 1 to round 2 which in line with other studies using similar setups ( Dufwenberg et al., 2005;Haruvy et al., 2007 ). Hence, the results show prices to be clearly above fundamental value in round 1 for both treatments. However, while in C the prices are close to the fundamental value in round 2, we still see severe overpricing in round 2 when margin purchasing is permitted in B .
To analyze treatment effects, we consider hypothesis testing and random effects regression based on the following model 1 .
Let the dependent variable X ir denote the measure of interest in cohort i in round r . The effect of the buy in margin treatment B is investigated by the dummy variable D B which takes value 1 for treatment B and 0 for C . The repetition dummy variable D R equals 0 in the first round when subjects are inexperienced and 1 in the second round when subjects are once-experienced. The product variable captures the interaction effect. Finally, v i indicates the time constant session specific error and ε ir indicates the idiosyncratic error term that varies over time and over sessions. Table 2 records the regression results for the bubble measures of interest.
Observation 1. Permitting buy on margin inflates asset prices.

Support:
The relative deviation is three times higher in B than in C in round 1, and six times higher in round 2, respectively ( Table 1 ). The regression 'RD' in Table 2 shows a significantly positive coefficient for the treatment dummy ( D B ). Also the positive deviation is significantly higher in B ('PD') as well as the highest price ('PMAX'). Neither of the interaction effects is significant, suggesting no change of treatment effect between rounds. Using session level data observation considering RD as the relevant unit, with eight observation in B and seven observations in C , we find a weakly significant difference in round 1 but no significant difference in round 2 ( p 1 = 0 . 064 , p 2 = 0 . 132 ). Experiences seem to have a dampening effect on prices such that first round differences in overpricing are not available any more in the second round. Table A.10 in the appendix provides p-values for treatment comparisons.
Subjects actively use the ability to buy on margin. To find out how many of the trades are debt financed, we count the number of trades in which a trader purchased a share and ended up/remained in negative cash holdings. In treatment B , about 20% of all transactions in round 1 were margin purchases, about 23% in round 2. From Table 1 we can see that average CREDIT equals 1857 Taler in round 1 and 1671 in round 2. To put this number into perspective, we compare it to the cash available at that time (sum of cash endowments plus sum of cash dividends). The ratio of debt to cash available was 0.25 in round 1 and even about 0.30 in round 2; a substantial amount of debt. JID: JEBO [m3Gsc;October 12, 2020;21:55 ]   Experimental results suggest that providing excessive liquidity in terms of cash leads to larger price bubbles (e.g. Caginalp et al., 2001 ). However, the literature so far neglected the endogenous inflow of cash via margin purchases. To test whether this inflow has an influence on overpricing, we consider the relationship between CREDIT and RD.

ARTICLE IN PRESS
Observation 2. Active margin purchasing has a reinforcing effect on overpricing.

Support:
The Spearman rank correlation coefficient between CREDIT and RD equals 0.762 ( p = 0 . 028 ) in round 1 and 0.929 ( p < 0.001) in round 2. Considering the number of executed margin purchases, we find the respective Spearman rank correlation with RD to be 0.786 ( p = 0 . 021 ) in round 1 and 0.585 ( p = 0 . 127 ) in round 2. The Spearman rank correlation coefficient between dept to cash available and RD equals 0.667 ( p = 0 . 071 ) in round 1 and 0.929 ( p < 0.001) in round 2. These results indicate a positive relationship between active margin purchasing and overpricing.
Margin trading is not without risk and our experimental results show that some traders bear too much of it. In four cohorts, margin calls were triggered. In these cohorts, the average number of margin calls was 5 in round 1 and 3.5 in round 2. In total, three traders ended up bankrupt following the leveraging of purchases on margin. To evaluate the question whether margin traders are compensated for bearing the increased risk, we classify traders into ' margin traders ', who purchased at least one share on margin, and others. We then calculate the difference of mean earnings at the end of a market between these two groups for each session and round. The average difference equals -110 Taler (SD 450) in round 1 and −163 Taler (SD 551) in round 2, i.e. margin traders on average earn less than others. However, using a sign rank test, we find no significant difference ( p 1 = 1 . 0 0 0 , p 2 = 0 . 674 ). 8 The received theoretical literature makes some suggestions about how the ability to leverage purchases can impact portfolio concentration. According to theoretical considerations of Fostel and Geanakoplos (2008) and Miller (1977) and experimental evidence provided in Carlé et al. (2019) , the most optimistic traders hold the largest number of risky shares in their portfolio when given the opportunity to do so. As treatment B provides sufficient liquidity for the 'optimists', we would expect an increased concentration of shares. We use two measures to test this prediction. First, we sum up the number of shares of the two subjects with the highest number of shares in their inventory. In round 1 (2) the average equals 10.93 (11.35) in B and 10.01 (10.10) in C . Using a Mann Whitney U test, we find a significant difference ( p 1 = p 2 = 0 . 037 ) suggesting a higher concentration when margin purchases are permitted. Second, we look at the coefficient of variation (CV) of stock holdings at the most extreme situations in each market. We find a higher CV in B than in C in round 1, but a reversed result in round 2 (see A.8 in the appendix). However, in both rounds we find no significant difference. Hence, we cannot definitely conclude that a higher stock concentration exists in B .

Experimental implementation
The reported results so far have been gathered under the condition that short sales are prohibited. In the second experiment, we permit short sales as a robustness check of the results on margin trading. The corresponding treatments are designated BS (buy on margin with short sales) and CS (control with short sales).
Traders make use of short sales when they sell shares without holding them in their inventory. In the margin account, cash serves as collateral on short sales. Short sellers assume a price decline to buy back the shares with a profit at a lower price. The amount of debt equals the market value of outstanding shares, i.e. debt fluctuates with market price (in contrast to margin purchases with a fixed debt amount). When the price exceeds a certain threshold, such that debt exceeds 50% of the cash, a margin call is triggered. 9 Immediately, i) the trader's buttons are disabled, ii) outstanding orders are cancelled, and iii) the computer automatically quotes a limit bid at the current market price. The computer continues buying back shares to cover short positions until the margin requirements are met again or until all short positions have been covered. Thus, in contrast to earlier studies, we do not restrict short interest and even allow for a short squeeze.
Short sellers pay dividends for their short positions at the end of each period such that a new period can already start with a margin call after the realisation of a high dividend. After period 15, both long and short positions are worthless. Note that margin traders can either margin purchase or short sell in treatment BS but only short sell in treatment CS . Since outstanding positions are closed even in case of bankruptcy, our implementation of the ability to short-sell shares differs to other asset market experiments, in which bankrupt short sellers are not able to close their positions and, thus, the share count is actually increased. 10 Apart from the ability to short-sell assets, the experiment has the same parameters as in B or C . The total number of subjects in the eight BS sessions and the eight CS sessions was 144 students. No subject had ever participated in asset market experiments before. The experiments were conducted at the MaXLab of the University of Madgeburg and at the BonnEconLab of the University of Bonn. Including comprehension questionnaires and payment, the session lasted up to 3.5 h. The sessions paid on average 30.44 euro in BS (SD 10.25) and 31.04 euro in CS (SD 10.82) including show-up fee.
Permitting the ability to short-sell enables pessimistic traders to express their opinion in the market, leading prices to reduced price levels ( Miller, 1977;Harrison and Kreps, 1978 ). In absence of the ability to short-sell, a situation arises in which optimistic traders alone determine prices when pessimists have run out of shares. Scheinkman and Xiong (2003) suggest that, due to overconfidence, heterogeneous beliefs create an opportunity to sell at a higher price to others in the future when short-selling is prohibited. Hence, we expect that the ability to sell short reduces overpricing which already has been confirmed in other asset market experiments (e.g., Haruvy and Noussair, 2006 ). The question is though, whether the ability to buy on margin still inflates prices when short selling is permitted. Therefore we are going to investigate our primary testable hypothesis also under the ability to short-sell assets; we conjecture that asset prices are higher under the treatment BS in which margin purchase is permitted relative to treatment CS in which margin purchasing is banned.

Experimental results
Analogously to Section 2.2 , we compare the measures of interest between BS and CS . Fig. 2 depicts the median period prices together with the fundamental value for both treatments for round 1 and round 2. In treatment BS we find overpricing to be substantial in both rounds, less so in round 2 though. The visible deviations in CS , in contrast, are small already in round 1 and disappear further in round 2. Apparently, overpricing is higher in BS than in CS . However, the ability to short sale in BS does not seem to eliminate overpricing, i.e. the price-inflation forces from margin purchasing are stronger than the price-deflating forces from short selling. Table 3 reports the average outcomes of the above defined measures for each round and also the differences between rounds. Overpricing in BS , as measured by RD, is about 51% in round 1 and still 31% in round 2-significantly positive in both rounds ( p 1 = 0 . 012 , p 2 = 0 . 012 ) but not significantly lower in round 2 ( p = 0 . 161 ). Overpricing in CS is much lower, about 7 percent in round 1 and about 0 percent in round 2, and not significantly different from zero ( p 1 = 0 . 208 , p 2 = 0 . 674 ). For RAD and for PD we see an insignificant reduction in both treatments and the average of the measures are higher in BS than in CS . The peak prices do not change so much from round 1 to round 2 but are higher in BS than in CS in both rounds. Again, the magnitude of overpricing is reduced from round 1 to round 2 in both treatments. The results indicate that buy on margin inflates prices even with short sales. Finally, prices in CS are not different from the fundamental value which is in line with the literature when short sales are permitted.

ARTICLE IN PRESS
The dependent variable is the measure of interest. The dummy variable D B (Buy on Margin Dummy) equals 1 if margin purchasing is permitted, D R (Round Two Dummy) is a round dummy and equals 1 for round two, and the product variable (Interaction) indicates the interaction effect between experience and the permission of margin purchases. Finally, v i indicates the time constant cohort specific error and ε ir indicates the idiosyncratic error term that varies over time and over cohorts. Parentheses record robust standard errors, clustered on independent sessions; * * * p < 0.01, * * p < 0.05, * p < 0.1. Definitions of measures can be found in the Appendix A.1 .
Observation 3. Permitting buy on margin inflates asset prices even if short selling is permitted.

Support:
Applying the same analysis as in Section 2.2 , we find that all measures RD, RAD, PD, and PMAX are significantly higher in BS than in CS (see Table 3 ) in both rounds. The regression analysis in line with model 1 reported in Table 4 also show a clear effect as the coefficients in the columns RD, PD and PMAX are significantly positive. Using a Mann Whitney U test with market level measures, we find a significant difference for round 1 in all measures ( p ≤ 0.024) and still some support for a difference in round 2 ( p ≤ 0.059). JID: JEBO [m3Gsc;October 12, 2020;21:55 ]  Again, subjects actively use the ability to buy on margin. In treatment BS , about 18 percent of all transactions in round 1 were margin purchases, about 24% in round 2. From Table 3 we can see that average CREDIT equals 1991 Taler in round 1 and 1784 in round 2. To put this number into perspective, we compare it to the cash available at that time (sum of cash endowments plus sum of cash dividends). The ratio of debt to cash available was 0.29 in round 1 and about 0.27 in round 2; again a substantial amount of margin debt.

ARTICLE IN PRESS
However, in contrast to the results without short sales we find no significant correlation between active margin purchasing and overpricing.
Similarly to the above reported results without short selling, margin calls were issued in four sessions and three subjects went bankrupt on their margin purchases in BS . There is no indication of higher or lower earnings of subjects who use leverage in their purchases. 11 Again, shareholding concentration, as measured by CV, seems to be higher in BS but we find no significant effect (see Tables A.9 and A.11 in the appendix).
To conclude, our data suggest that the ability to purchase assets on margin inflates prices even if short-selling is permitted.

Overview of results
Although our focus is on margin purchases, we want to briefly combine selected results in a classical 2x2 design overview. Table 5 shows the differences in treatment averages for the relative deviation (RD) and the highest price (PMAX). Additionally, we report p-values for a Mann Whitney U test with seven ( C ) or eight observations per treatment. The tables for all measures can be found in the appendix ( Tables A.10 and A.11 ).
As discussed above, margin purchases increase overpricing by about 50% without short selling and 44% with short selling. The average highest price is more than 200 Taler higher without short selling and more than 100 Taler higher with short selling. The table thus provides an overview of the results we have discussed above. In line with the literature (e.g. Haruvy and Noussair, 2006;Ackert et al., 2006;Weitzel et al., 2020 ), we find lower overpricing in markets with short selling but also lower maximum prices ( B vs BS , C vs CS ).
We find the highest differences when comparing B and CS , which is of course no surprise as the measures for B are higher than for C and the measures of CS are lower than for C . If the dampening price effect of short selling and the enhancing price effect of margin purchases neutralized each other, then we should observe that the difference between C and BS should be about zero. However, we find the relative deviation and the maximum price averages to be lower in C than in BS , significantly so in round 2 only. These results suggest that the price effect of margin purchasing is stronger than the price effect of short selling, at least for our experimental implementation of the two institutions.

Discussion and conclusion
For the first time, we have tested the effect of the ability to buy on margin along with the threat of margin calls and forced liquidation in an experimental asset market environment. In contrast to other studies, the purchasing power of the JID: JEBO [m3Gsc;October 12, 2020;21:55 ] market participants endogenously expands or tightens when asset prices move higher or lower. Again, laboratory experiments serve as a test-bed for market regulation designs which cannot be tested in real-world exchanges without implementing an expensive natural experiment ( Bloomfield and Anderson, 2010 ). Our main results are quite intuitive and in line with the theory. With margin purchases, we observe price inflation relative to the baseline. These qualitative results are unaffected when short selling is permitted. Leverage on individual portfolio decisions seems to increase risk-taking characterized by higher concentrations of risky assets, eventually resulting in individual bankruptcies. Thus, our experimental results are in line with theories of margin purchasing by Fisher (1933) and by recent heterogeneous agents models, Geanakoplos (2010) that conjecture such effects on asset pricing and portfolio decisions.

ARTICLE IN PRESS
As in any laboratory experiment, the experimental design choice might affect the results. For example, Duchêne et al. (2019) implemented 'borrowing treatments' where traders had to pay back half of their endowment at the end of the market. Furthermore, they considered call markets where borrowing and short selling depended on the clearing price. In our setting, the market price determined the borrowing capacity, and traders explicitly chose to buy on margin. In fact, Füllbrunn et al. (2014) showed that feedback in dynamic trading environments changes the performance in comparison to feedback in static trading environments.
In a Wall Street Journal article "Margin Calls: Should the Fed Step In?" ( Shiller, 20 0 0b ), Shiller proposed an active margin regulation in times of significant market misalignment to mitigate the Minsky credit cycle (e.g., Nikolaidi and Stockhammer, 2017 ). Asset market experiments might help to understand margin requirements. In particular, testing time-varying margin regulation policies might inform policymakers regarding potential implementation strategies. Our contribution takes only the first step by considering a situation in which margin purchases are either permitted or not. Thus, we confirm that a ban on margin purchases significantly reduces overpricing. We suggest future research to take further steps by considering different maintenance margin amounts or overpricing-related margin policies

PMAX Highest price observed in a market SHORT
Highest number of short interest observed in a period CREDIT Highest amount of credit observed in a period

Measures 2
Geometric Deviation (GD) 15 Greatest number of consecutive periods that median transaction prices are above fundamental values Turnover (TO) 15

Measures 4
Shareholdings (   Notes: The table reports the difference of treatment averages together with the p -value of a Mann Whitney U test comparing the measures across treatments ( * * * p < 0.01, * * p < 0.05, * p < 0.1). Definitions of measures can be found in the text or in the Appendix A.1 .  JID: JEBO [m3Gsc;October 12, 2020;21:55 ] Appendix B. Instructions   Fig. A.1 shows a screenshot of the trading screen. On the left hand subjects were able to post or delete limit bids and limit asks. Information on demand and supply at each point in time were available. Traders were able to vote to end the period and a embedded calculator helps to make decisions. On the right hand information on the margin account were given, a history graph on recent prices along with a table with information (open, high, low, close, volume, dividends, cumulated dividends, shares in inventory, cash balance). Subjects can hold shares or exchange shares and cash in real time.

ARTICLE IN PRESS
To buy a share, subjects submit bids; to sell a share, subjects submit asks. Asks and bids are submitted as limit orders, stating the receivable or payable Taler amount for one share, respectively. Unless a trade occurs, orders are quoted in the open order book that is common information. Enhancing orders are listed first; non-enhancing orders are listed second. Hence, traders observe the depth at ask and the depth at bid. The difference between the top listed bid and the top listed ask is the bid-ask-spread. Orders are filled, i.e. a trade is executed when a new bid (ask) exceeds (undercuts) or equals the best offer to sell (buy). Upon submission of a matching order, a trade closes. The best outstanding order in the open book is accepted as the trading price and the quote of the order is eliminated from the open book. The seller's cash balance is increased by the amount of the price and the number of shares held is reduced by one. The buyer's accounts are treated correspondingly. Subjects can withdraw asks and bids from the open order book by deleting the quotes. Trading prices are displayed in chronological order in a table visible to all subjects.
The following instructions (translated from German) show instructions from treatment B. In most of the time, the wording is almost the same as in the other treatments and was adjusted where necessary. After the instructions you will find the main treatment changes in BS and M. MS includes both instructions. The German instructions as well as the questionnaire for training the institutions can be obtained on request. Comments are in brackets.

Instructions
Welcome to the experiment. Please read these instructions carefully. Each participant has the same instructions. If you have any questions, please address them to us. Do not speak to other participants and please stay quiet for the duration of the experiment. Please switch off your mobile phone now, leaving it switched off until the end of the experiment. If you dis-  obey these rules, you will be excluded from the experiment, losing any right to payment. This experiment is about trading assets. You own shares and buy or sell them in a market. A market consists of nine participants. During the entire experiment, you always negotiate with the same eight other participants, whose identity remains unknown to you. All transactions are in "Taler". All Taler that you earn are converted into Euro and paid to you in cash at the end of the experiment. The rate of exchange is 100 Taler = 1 €. In addition to the payoff at the end of the experiment, you will receive 4 Euro for show up. The experiment consists of two rounds of 15 periods each. Before the beginning of each round, there is a short practice round. In this practice round, you can acquaint yourself with the market situation and the computer program. Transactions in the practice round have no effect on payment. The experiment will last for about 2-3 h, including the introduction and the practice round.

Description of a Round
One round consists of 15 periods. In each period, you can buy and sell shares on the market. At the end of the period, the owner of a share is paid a dividend. The dividend is a sum of money that is credited to the owner and remains at his disposal for the rest of the round. The amount of the dividend is randomly decided (by the computer) at the end of each period. It is the same for every share: 60, 28, 8, or 0 Taler. Each of these sums is equally likely, i.e., the likelihood is one quarter (or 25%), independent of how high previous or subsequent dividends are. The dividend is on average, 24 Taler per period ( = 0.25 * 0 + 0.25 * 8 + 0.25 * 28 + 0.25 * 60), 48 Taler in two periods... and 360 Taler in 15 periods. The Table  B.1 recapitulates the sum of the average future dividend payments for each period.
The table can be read as follows: assume you are in Period 8 (column 1, line 9). Including the dividend that is due at the end of the period, there remain 8 dividend payments (column 2, line 9). Each of these dividends has a 25 % chance of being 0, 8, 28, or 60 Taler, and on average 24 Taler (column 3, line 9). The sum of all average future dividend payments is therefore 192 Taler ( = 8 * 24), as you can see in the fourth column of the table (column 4, line 9). In the most convenient case, if from the 8th period on wards a dividend of 60 Taler is paid each time, then the maximum possible dividend payments add up to 480 Taler, as can be seen in the last column of the table. However, it is also worth noting that the most disadvantageous situation could mean the sum of all dividends being 0. Please note also that the share becomes worthless at the end of the round, i.e. after the 15th dividend payment; there are no further payments (last line). After the last period, the depot is closed.

Market Organization and Trading Rules a) Conducting a Trading Period
In a trading period, you have the chance to buy and sell shares. In order to buy shares, you need money. On the righthand side of the screen, you will see the sum of money currently available to you. In order to sell shares, you need shares. Your depot shows your shares currently available to you. During a trading period, you may conduct the following actions (see Figure 1 I. Sale Offer: You can offer a share from your depot for sale. In the field underneath the button Sale Offer, you key in the minimum price you wish to receive for a share. Then, you confirm this minimum price by clicking the button Sale Offer. Your sale offer is then added to the list All Sale Offers, where all offers are placed in descending order, from the highest to the lowest. All market participants can see this list. If two or more sale offers are the same, they are listed in chronological order. Your own sale offer is displayed in blue, while the other sale offers are visible to you in black. In addition, the list JID: JEBO [m3Gsc;October 12, 2020;21:55 ] Own Sale Offers is established. Each sale offer stands for one share. You cannot offer more shares than you possess in your depot. II. Purchase Offer: You may put in a bid for the purchase of a share. In the field underneath the button Purchase Offer, you key in the minimum price you wish to pay for a share, i.e., the maximum sum you wish to pay. Then, you confirm this maximum price by clicking the button Purchase Offer. Your purchase offer is then added to the list All Purchase Offers, where all offers are placed in ascending order, from lowest to highest. All market participants can see this list. If two or more purchase offers are the same, they are listed in chronological order. Your own purchase offer is displayed in blue, while the other purchase offers are visible to you in black. In addition, the list Own Purchase Offers is established. Each purchase offer stands for one share. The sum of your purchase offers may not exceed your assets.

ARTICLE IN PRESS
III. Deleting Sale Offers and Purchase Offers: You may retract a purchase offer by double-clicking a purchase offer in the list Own Purchase Offers, marking it and deleting it. You may retract a sale offer by double-clicking a sale offer in the list Own Sale Offers, marking it and deleting it.
Conclusion of a Trade: A trade is concluded as soon as an offer to sell is lower than or as high as an offer to buy. The price at which the trade is executed is determined by chronological order: 1. if a new sale offer is lower than or as high as the highest existing purchase offer, the trade price is the same as the highest purchase offer. 2. If a new purchase offer is higher than or as high as the lowest existing sale offer, the trade price is the same as the lowest sale offer. If a sale is executed, the transaction price is entered in the list of trading prices, and the purchase offer of the buyer and the sale offer of the seller are deleted from all lists. The depots of the buyer and the seller are automatically updated: The seller's depot is lowered by one share, and one share is added to the buyer's depot. At the same time, the trade price is subtracted from the buyer's account balance and added to the seller's account balance. Please note the trade restrictions: you may only buy a share or make a purchase offer if you have enough cash at your disposal. You may only sell a share or offer it for sale if it is in your depot. You may not sell shares to yourself. Your sale offers must lie above your purchase offers.
Insert further instructions from other treatments here A trading period lasts 180 seconds. The remaining time is shown at the top right of the screen. If you quit your trading activities before the end of a trading period, you may activate END by clicking on the corresponding field with your mouse and then pressing ENTER ( Fig. 2 ). After that, you can still trade, but the period ends prematurely if all traders have simultaneously clicked on END. If you wish to change your decision, you may activate CONTINUE (this is the default setting) by clicking on the corresponding field with your mouse and then pressing ENTER. In that case, the trading periods only ends when the time is up. Figure 2: Ending trade prematurely [ skipped ] b) Information at the end of the period A detailed account follows each trading period. You are informed of the dividend achieved; the total proceeds from the dividend and your new balance. An overview of the period accounts can also be found in a table on your screen. Here, you will see the opening price of the period, the highest price, the lowest price, the final price, the number of sales agreement contracts, the amount of the dividend, the income from the dividends, the number of shares in the depot and the balance at the end of the trading period. In addition, you will find, on the left, a graph showing the prices from the past periods. The graph also shows the opening price (green) and the closing price (red), as well as high (black, above) and low (black, below), so that a chronology of the past trading periods can be drawn up. The account is visible for only 20 seconds. You will also find the graphs and the table during the trading period. To continue more quickly, click Continue.

Initial Endowment
A market consists of 9 participants. At the beginning of each round 18 shares will be provided to the participants. The allocation of shares is due to a random process. Each participant is endowed with money and shares at the beginning of each round. There are 3 possible endowments:

Endowment Class
Money Share   I  225  3  II  585  2  III  945  1 3 participants receive endowment class I, 3 participants receive endowment class II, and 3 participants receive endowment class III.

Initial Endowment
The experiment procedure is as follows: • Practice round for a trading period • Round 1 : Allocation of initial endowment, 15 (Trading periods + Dividend payments) → Account balance after Round 1 .
• Round 2 : Allocation of initial endowment, 15 (Trading periods + Dividend payments) → Account balance after Round 2 . JID: JEBO [m3Gsc;October 12, 2020;21:55 ] In other words, the same rules count for each round. At the beginning of each round, the initial endowments are allocated via the distribution process. The balance from the first round is no longer at your disposal in the second round, but is instead paid to you at the end of the experiment along with your second-round balance. The account balance in each period consists of the following elements: Initial endowment according to the distribution process + Income from dividend payments + Income from the sale of shares -Expenses relating to the purchase of shares

ARTICLE IN PRESS
The sum you are paid in Euro at the end of the experiment is calculated as follows: (account balance after Round 1 + account balance after Round 2) * 0.01 Euro + X Euro.

Instructions for selling short [Treatments BS and CS ]
Borrowing shares : You are able to sell shares without possessing them, i.e. you lend a share to sell it (Short Sale). A short sale increases your cash account. It also increases your debt. As long as you do not cover the short sale with buying a new one your debt exists. During having short positions your debt equals your outstanding shares evaluated by the current market price, i.e. your debt varies with the trading price. You equity equals your cash minus the value of your short positions (debt). Your available funds equal your equity minus your debt, i.e. your debt cannot exceed your equity. For each borrowed share at the end of a period you have to pay the dividend. Your dividend reduces your cash and, thus, your equity. With decreasing prices, your debt decreases because the value of shares that you have to buy back decreases. With increasing prices the value of shares increases and, thus, your debt as well. If your debt exceeds your equity due to a price increase or due to dividend payments, you have no more available funds. The price threshold at which your debt exceeds your equity is provided at your screen. When the price exceeds this price threshold, all your current offers are canceled and an offer to buy at the current market price is automatically submitted. The purchase will decrease your debt. If afterwards your debt still exceeds your equity, this process will repeat until your equity exceeds your debt again. Then you are allowed to submit or delete orders. After the final period, shares remain worthless. When at this point you have borrowed shares outstanding you automatically cover them for a zero price. During trading periods you will be informed about your available funds, your cash at hand, the value of your shares at current market price, your equity, your debt and the price threshold. Why should you borrow shares? When you buy a share later for a lower price that you received now, you make a profit that equals the price difference. However, note the dividend payments at the end of the period. Thought selling short can lead to higher profits when prices decrease. When prices increase or you have to pay dividends, also losses may occur.

Instructions for margin purchasing [Treatments CS and BS ]
Available funds : You can dispose of equity and debt. Your equity equals the sum of your cash at hand and your shares in your inventory evaluated by the current market price (stock deposit value). If your cash at hand is not enough to buy shares, you will receive additional funds without any interest payment (this is negative cash). You can borrow money as long as debt does not exceed equity. Your available funds to buy further shares equal your equity less your debt. Thus, your share inventory can be finance with debt equaling half of the value of your inventory. Borrowed money has to be returned until the end of period 15. Equity and available funds vary with prices. Debt is reduced (and cash is increased) with each dividend payment. An increase in market prices increases your stock deposit value and, thus, your equity and your available funds. A decrease in prices reduces your equity and, thus, available funds. When your debt due to a fall in prices exceeds your equity, you have no available funds left. The price threshold at which debt exceeds equity is provided at the screen. A drop in prices below this threshold will cancel all your current offers and an offer to sell at the current market price is automatically submitted. The purchase will decrease your debt. When your debt still exceeds your equity after the forced sale, this process will repeat until your equity exceeds debt again. Then you are allowed to submit or delete orders. During trading periods you will be informed about your available funds, your cash at hand, the value of your shares at current market price, your equity, your debt and the price threshold. Why should you borrow money? When you sell a share later for a higher price that you paid now, you make a profit that equals the price difference. Thus, purchasing shares with additional money can increase your profit. Thought borrow money to buy shares can lead to higher profits when prices increase. When prices decrease also losses may occur.