The following is a lightly edited version of an item initially posted on my blog at Knowledge Problem: Secrets of an Inkling Top Trader: Spotting Riskless Arbitrage Opportunities:
As mentioned previously at the Knowledge Problem, Inkling offers a public play-money prediction market. I stumbled across them a year or so ago, and because I’m interested in market design and prediction markets, I decided to try them out. Partly because playing the Inkling markets amuses me and partly because I started doing well, I’ve continue to play on Inkling. Eventually I wormed my way onto their Top Traders list, where I’ve remained for several months.
In the process of amassing my play-money riches, I’ve learned a few useful things about Inkling’s markets. Here I am “giving back” to the Inkling user community by sharing one of my secrets.
This is it: when a market offers you free, riskless profits, take them.
Obvious, right? The trick is in spotting the riskless profits. I recently was able to take some free, riskless profits when Inkling allowed two markets to be set up for the same event: the UEFA Champions League. It isn’t necessary for there to be two markets on the same event for arbitrage to work – I did something similar in the NCAA men’s basketball final four market – but the two market case makes the arbitrage process easier to understand.
By the way, as I write the markets are still live, and I have existing risky holdings in these markets. These stakes are standard bet-I-know-better-than-the-market plays which may or may not pay off. None of what I am about the explain involves taking risks, just taking profits.
The standard multiple result market at Inkling pays off at 100 units for each share of the winning outcome you hold at the closing of the market, and pays off at 0 for all other outcomes. In this case both markets will pay 100 in play-money units for picking the eventual winner in the 2006-07 UEFA Champions League, and 0 for shares in all other outcomes. Because two separate markets are running over the same event, when the prices of the markets diverge, there is a simple and obvious arbitrage opportunity. Of a pair of matching outcomes (Say “Chelsea wins”), buy the lower priced of the two and sell short the same number of shares in the higher priced market.
The final outcome doesn’t matter, because whatever you win on one you will lose on the other. Your entire profit is captured in the net proceeds of the buy low-sell high pair of transactions. Here is a simple example drawn from my trading report:
| Market | Outcome | Action | Proceeds/Cost | Date – Time |
| 1st Market | AC Milan | 50 sold | $685.82 | Apr 30, 2007 – 15:09:28 PDT |
| 2nd Market | AC Milan | 50 bought | -$545.71 | Apr 30, 2007 – 15:09:53 PDT |
| Net: $140.11 |
I sold into the first market at an average price of about $13.72 and bought in the second market at an average price of $10.91. The result: a $140.11 profit for a minute or two of effort. Not a lot of profit, but not too bad for riskless trading. (The short sale does tie up some credit, but since we’re dealing in a play-money exchange that doesn’t pay even play-money interest on your holdings of pretend cash, the only thing you sacrifice is other trading opportunities.)
Outside of the rare pair of auctions covering the same event, I’ve found similar riskless trading opportunities in the recent market to pick the final four teams in the NCAA men’s basketball tournament. The market would pay 100 for each of the four teams that reached the final four weekend. The market started even before the tournament seedings were announced, with a large number of possible teams, and shares traded on various market expectations. Curiously, early on the total value of all of the stocks summed to more than 400 even though owning one of each share was guaranteed to pay off exactly 400. (This is a clue that arbitrage opportunities are available.)
Once the tournament seedings were announced, it is clear that owning one of each share of all 16 teams in a single region would payoff exactly 100. If the share prices summed to less than 100, then buying one share of each team generates a payoff at the difference between the cost of the shares and 100. If the share prices for all teams in a region sum to more than 100, then selling one share each will generate profit to you at the difference between the proceeds of the sales and 100.
For simplicity, say that there are just two teams left in the “West Region.” Let’s call them, hypothetically, “UCLA” and “Kansas.” Buying one share of UCLA and one share of Kansas, will lock in a payoff of 100, so if the prices of the two teams sum to less than 100 then you can obtain riskless profits by making the purchase. (Or sell short if the prices sum to more than 100.)
While it may be less obvious, the underlying arbitrage is similar to the AC Milan example. At this stage of the contest, a share of “UCLA wins” is the logical equivalent of “Kansas loses.” Buying one share of UCLA and one share of Kansas is the logical equivalent of buying a stake in “Kansas loses” and a simultaneous stake in “Kansas wins” – with offsetting holdings your payoff doesn’t change based upon the outcome of the event, and your profit is risklessly captured from arbitraging the market at the time of the transactions.
Many Inkling markets are self-arbitraging in the sense that they automatically account for these interrelationships in pricing multiple outcome prediction markets. For example in the separate market to pick the eventual champion of the NCAA men’s basketball tournament, all prices automatically adjusted in response to any purchase or sale such that the sum of the prices always totaled exactly 100. (In fact, both of the two UEFA Champions League markets are self-abitraging within the markets, but I profitted by arbitraging between the two markets.
As an economist, my opinion is that such self-arbitraging markets likely exhibit superior efficiency properties that would make them desirable in real-money practice. (As a sometime experimentalist and aspiring prediction-market geek, I’d love to test that conjecture in an econ lab.) As an Inkling trader, however, I love to discover and exploit riskless trading opportunities in non-self-arbitraging markets.
[Okay, I’ll admit that I didn’t become an Inkling Top Trader via riskless arbitrage. I took many risky steps along the way, some of which paid off handsomely. But explaining that I often got lucky doesn’t appeal to my inner aspiring PM-geek.
BTW, in addition to pulling riskless profits out of the UEFA Champions League markets I was also carrying very substantial risky holdings in both Chelsea and Man U. Surely one or the other would win it all, right?
Ouch.]
Did you prove that 1) you’re really smart or 2) that it’s easy to game a market which has no consequence and is fundamentally illiquid / inefficient?
This is like beating a 7-year old at a math test. Congratulations.
I’ll say it again: play money markets’ predictive power is inferior to that of real money markets. Play money markets do not approximate the real world.
I do believe play markets can absolutely approximate the real world — but they need genuine, participation. These 20-30 person markets are about as useful as a show of hands in a class room debate.
In the real world, markets are never truly efficient. They can never be 100% liquid.
The closer you get to 100% liquidity, you pay lower transaction costs, but there are lower incentives for price and information discovery as well.
Play-money markets aren’t just 100% liquid; with automated market makers and other artificial liquidity boosters, “self arbitrage” etc, they’re over 100% liquid. They can break that law because liquidity is “free” in play money markets — but that means that the underlying prices/ probabilities have that much less validity. (BTW, self-arbitraging real money exchanges would go bankrupt. It’s an awful idea.)
So not only are people not really penalized for leaving stupidly mispriced offers, aka arbitrage opportunities, out there (we all saw that coming) — but people can also take advantage of AMMs to squeeze free profits as well.
These facts are not good for play money predictive power.
QUOTE I do believe play markets can absolutely approximate the real world — but they need genuine, participation. These 20-30 person markets are about as useful as a show of hands in a class room debate. UNQUOTE
From memory, the MSR prediction markets need 12 traders —see Jed Christiansen’s paper.
Mark, it doesn’t seem to me that the arbitrage is “gaming the market,” but I suppose the term doesn’t have a precise definition so maybe by your definition. I agree that it is easier to find this kind of arbitrage offer in play money markets, but the point was to illustrate the principle in the context of the AMM mechanism that Inkling uses.
The point wasn’t to prove that I could, you know, probably beat a 7-year old at a math test. Perhaps you’re reacting to the somewhat didactic tone. The piece was written with a non-specialist audience in mind (e.g., regular KP readers and Inkling users), but Chris Masse invited me to cross post it here.
Alex, I didn’t make any claims about the predictive power of play money markets. You may or may not be right – I haven’t collected any evidence and can’t give you an informed response.
Michael, your piece is very interesting. As you see, many real-money event derivative traders are skeptical when it comes to play money. But this will fade out as more evidence about their accuracy surface.
Thank you for coming here, on Midas Oracle, where we like to have a variety of authors and topics. And that includes Barry Ritholtz, who is skeptical about *real-money* prediction markets.
http://www.midasoracle.org/2007/02/21/misunderstanding-prediction-market-failures/
Alex, Your more interesting point (to me) concerns liquidity.
I’m not sure what counts as an “artificial” liquidity booster (or the complementary set – the “natural” liquidity booster), but it seems to me that it is well worth exploring market designs that facilitate liquidity. I know that Ledyard and folks at Caltech have investigated this question in the lab (though in other contexts, not with Hanson’s MSR).
Wouldn’t it make sense that improving liquidity would give the underlying prices/predictions more validity, not less? Poor liquidity just results in some information not coming into the market.
With respect to your remark about “stupidly mispriced offers,” I think one way to frame the story of financial innovation in the 1980s-1990s and continuing today is as a search for mispriced opportunities to provide liquidity to the market (see Richard Bookstaber’s recent book for example). Some of the quants on Wall Street made their money in pure arbitrage (in real money exchanges like the NYSE) and others made money by finding lower-risk ways to provide liquidity to the market.
It may be true that a self arbitraging real money exchange based upon automated market makers can be made to work as a money pump. I think this could be because the arbitrage is not perfect across all dimensions in multi outcome markets – the MSR eliminates the kind of gross arbitrage described in the post by ensuring prices in the market sum to the payoff, but there is no guarantee that the MSR will allocate the price adjustments needed in a way the reflects all implicit in the information that motivated the initial transaction. There may in fact be some residual risk in the way the arbitrage is implemented by Inkling (albeit they are “risking” play money), or maybe the risk is inherent in the AMM design.
But I absolutely disagree that self arbitraging real money markets are a bad idea in general. When the market design can facilitate truly riskless exchanges, doing so enhances the efficiency of the market.
In the case of “Does Money Matter?” the information being aggregated was already discovered and dispersed, but as the authors add, “Theory suggests that real money may better motivate information discovery”.
http://www.newsfutures.com/pdf/Does_money_matter.pdf
Liquidity not backed by information may result in prices less useful than having no price. Perhaps confidence bands on prices can be established based on the number of real traders participating in a market.
Implementing auto-arbitrage makes sense although some sites opt against it in order to encourage more participation (but the arbitrage may not be realized if there are many markets, if the arb opportunities swamp real-trader liquidity).
Jed,
I have also read the SSRNs that argued in favor of play money markets. I remember we had a pretty good debate here several months ago on the very subject, and it ended with a big fat “Unresolved.”
Michael,
No, I adamantly disagree that liquidity should be artificially “facilitated.” There are certainly circumstances in which an exchange might want to temporarily boost liquidity, if they think they have a good opportunity to increase long run trader population… but… Liquidity of a security, in the long run has a lot to do with information awareness relative to other securities. I.e., if traders are staying away from it, that means they are finding easier alpha elsewhere, and the alpha in that particular security, relative to the information discovery/aggregation cost, simply is not worth it to them. If they don’t have a significant alpha edge on the security in question, they shouldn’t trade it.
That was a very crappy explanation. But liquidity generally shouldn’t be subsidized. Just as inflation isn’t good for the market just because it promotes more trading. There is more marginal trading/liquidity, but that is significantly outweighed by inflation’s mispricing effect. (I wish I had more time to give a more thoughtful reply to your question.)
Cheers,
Alex
Jason beat me to it, for the most part.
Although I am not a fan of self-arbitraging mechanisms. Those can only arbitrage based on prior information, not current information. During short time intervals of high volatility, when incoming information has more weight than historical information, auto-arbitraging mechanisms are begging to be exploited.
I wonder if the different trader populations on the two sites biased results one way or the other. What % of Newsfutures traders were based in Europe vs. Intrade, for instance?
If play-money markets sometimes reveal information faster than real-money markets, I don’t think they do so on average, thus making it hard to use that fact. Real-money markets should be closer to risk neutral than play money markets so i would guess they will do better on average.
The Cowgill lottery system helps play money markets. It might also be feasible to calculate the optimal risk preference based on the prize structure and use that value to back-out risk neutral probabilities.
We debated the NF/Intrade contrast at length. Judging from Intrade’s forums, its traders both 1) tilted very right of center, and 2) did not do much apparent information aggregation.
The point of a market is to have predictive power. Although the play-money markets in that instance were, on average, closer to the final outcome, NF was also orders of magnitude more volatile.
What does a “probability” of 50% mean if it could well be at 70 percent a few days from now, on the basis of a couple of pretty insignificant marginal data, as in NF’s case? You could make an argument that play money markets react more quickly to marginal data, but they also grossly overreacted to it.
Also, to second Jason again, the NFL faceoff between Intrade and NewsFutures concerned information already aggregated. Everyone was trading on the same data; the choice of subject matter was inherently flawed.
As far as I can tell, the play-money markets are much closer to “conventional wisdom,” which vastly overestimates current information relative to historical information, and fluctuates way too much.
NEWSFUTURES: I wonder whether the French traders, who are anti-Bush, were responsible for the Bush re-election event derivative being a bit less priced than on TradeSports-InTrade in October and November 2004. NewsFutures predicted Kerry in the end, after many hesitations. As for the French presidential election, I’m happy with NewsFutures: the prices were not too much far away than InTrade and BetFair, though these two real-money prediction exchanges command more respect since their two prices were quite similar.