As noted in this post, I’ve been teaching an Order and Execution Management class at work. In my attempt to make this class interesting and interactive, I’ve supplemented lectures with trading games. For the first session, it was bartering animals; for the second session, it was acting as a market maker and managing a limit order book based on the suggestions in Clark and Mary-Carol Maxam’s white paper on Efficient Markets and Information Processing.
If anyone would like to attempt recreating this game, you can use the spreadsheet available here.
How it Works
For this exercise, I divided the class into teams, with each team acting as a market maker in a synthetic asset; the intrinsic price of this asset would be the average weight of everyone in the class. After collecting the average weight of each team, I explained: imagine we’re specialists in trading a security, and the real price of that security is the average weight of everyone in the class. As market makers, each team needs to set a bid and ask price that straddles their estimate for the value of this fake synthetic security. The bid and ask spread must be 10 lbs. For example, if I thought the average weight was 165 lbs, I might set a bid of 160 lbs and an ask of 170 lbs. Each team thus had partial information about the value of this asset, and no one was more informed than any other.
The Limit Order Book
I gave teams 90 seconds to come up with bid and ask prices, and then put those prices into a spreadsheet and converted into a limit order book (LOB) for display to all. As we can see here, folks had a very divergent set of expectations about the value of the fake asset. Further, this book was crossed; that is, the best bid was higher than the best ask:
After displaying the LOB, each team had the opportunity to buy 1 share of this asset, or sell 1 share of this asset. They could only do one or the other, and could only transact in 1 share. Buying and selling had no impact on the quotes displayed in the book; if someone sold a share to venue 5 at 165 in the above example, that wouldn’t clear the bid – rather, the bid would remain in place for the entire round of trading. In this manner, each team had the same opportunities as each other. Trading happens in public, so team learns what other teams are doing in turn.
After round 1, teams then have another 90 seconds to revise their bid and ask prices (amongst themselves, privately), we enter those prices into the spreadsheet, regenerate the book, trading commences, and so on for a total of 10 rounds, ultimately generating this set of bid-ask spreads per team over time:
The Results
What we see is a lot of variance for the first few rounds, and eventually the bid-ask converge and remain straddled, more or less, around the true asset price (the average class weight in pounds). What we see is that without any explicit new information about the true asset value, prices changed and converged correctly. As each team came to understand how other teams viewed the true asset value, spreads converged and the focus of the game moved to one of inventory management.
Rather than attempting to find correlations, the objective here was to internalize the flow of managing the book, balancing inventory management alongside price setting and discovery. A second goal of the game was to maximize P&L. At the end of the game, we calculate P&L by marking to market inventory changes against the true price. For example, if the average weight of the class was 165, and in round 1 I bought a share at 170, I’d have a net P&L of -5. By balancing inventory, teams could attempt to manipulate themselves into a positive P&L.
Final Round Review
In the final round, rather than marking ending inventories to market, however, teams were forced to liquidate or cover inventory at the final best bid and best ask prices, as if they were submitting market orders to get out completely from this market. For example, if a team had 5 shares at the end of round 10, and the best bid was 152, then all 5 shares would be sold at 152 and P&L for each would be calculated against the actual asset value. We can see the results of the game in this chart, which shows cumulative inventory and P&L for each team over time (note that this chart excludes the results of the forced market value orders executed following round 10):
We can see that team 3 made some bad decisions early on, and was unable to recover from the early negative P&L.
This chart shows final P&L per team, incorporating the forced market orders executed at the end of round 10:
