Start a new topic

Fee structure distorts the market

Hi, others have alluded to this, but I haven't seen anyone come out and say it directly.  The current fee structure prevents this market from functioning properly.  Ask any economist, if you have a 10% take on profits and a 5% withdrawal take, you are going to influence behavior.  In this case, it has the effect of preventing most arbitrage, because gross spreads on trades have to be extremely large for it to make economic sense to put money on this site to do an arbitrage.


I get that the site is non-profit, and there are real costs to running it (for the record, I would not be against a profit for the site).  That said, I don't understand though why there isn't a subscription option for users, as it would eliminate the effects the fees are having on the market, which would create more realistic odds (assume that is academic purpose?) as well as a better experience for users.


A few examples specific examples:


a) The current Republic nomination odds are a joke.  They add up to well over 100% (top 7 alone add up to 139% on the "sell-yes" side of market right now, which is the easiest way for me to think about it, all 21 add up to something like 180%).  Let's say I sold 850 shares of each of top 7, that would cost me $4,769.  If none of them wins (not impossible), I make $1,181, or 25% gross.  More likely, one of them wins, and costs me $850, so I really make $331.  That's still 7% gross, and a 10% IRR to the republican convention.  That's a big spread.


In reality though, I'm going to have to pay ~$100 fee on my gross winnings depending on who wins (10% of my gross and I can't deduct the $850 loss I had to take to get here!).  I also have to pay 5% to take the capital back off the site, assuming I didn't want to let it all ride after the arbitrage (that's another $250-ish based on my 4,768 + ~$230 gross gains).  That is all my winnings.  I have a 7% guaranteed gross trade, and I can't make any money on a net-basis.  This is why the odds don't make any sense on the site, because the fees prevent arbitrage. 


b) Just to point out that this isn't the only place where such a discrepancy exists, Hillary + Bernie odds for winning the general election right now are 69% on the offer side, despite democratic odds of winning presidency at only 61%.  This is a huge spread for a normal market, but when you consider you'll be taxed on profitable trades, without an offset for losses, it's not very interesting.  


I could go on - you will notice nearly every question has odds that currently add up to well over 100%.  No one can take advantage of this because no one wants to add too much capital to the site since they're going to get hit with a 5% cut across the board to remove it later.  


Again, I want to be clear I am all for the site making money, but how it is structured today is influencing the odds and distorting the market in obvious and illogical ways.  For sake of argument, let's say site was willing to allow users to opt in to an alternate $350/yr subscription rate as the only fee (which seems very high for most users, but is in line with the fees generated from the trade above).  Who wants to bet that the spreads above would collapse within days?  I think it would meaningfully improve efficiency/accuracy of the market odds, and would still allow plenty of room for a very nice business margin/model for the operators and I'm guessing most users of the site would happily opt in at a more reasonable price point (maybe ~$100/yr?) although I'd love to hear others' thoughts or opposing viewpoints.




19 people like this idea

I agree entirely with gk123's point.  It seems to me that Predict It has an academic / research mission of developing a prediction market to tap the wisdom of crowds, combined with the discipline of putting real money at risk, to the ultimate goal of getting the best available estimates of various probabilities.  That goal has been thoroughly undermined to date by (a) the lack of linked-contract margin which makes arbitrage require a very high ratio of capital to profit and (b) the illogical fee structure which makes arbitrage an unprofitable activity except at absurdly high spreads.  Either a subscription model like gk123 lays out (though I would hope $1-2 per month would be sufficient) or a per-contract-traded fee ($0.001?) would work, or the greater of the two to effectively penalize inactive accounts and allow the site to have sufficient revenue when activity inevitably falls off a bit post election.  This would allow PredictIt to be much more successful in achieving its own goals.


8 people like this
I don't think this is a good idea. monthly fees would turn away a lot of new potential users. also linked margin will fix most of the problem with overpricing. Just look at the Obama weekly market, bids rarely sum to 100% and if you ever get to 108% people start hitting out bids to arb. Running the site on a monthly fee would turn away almost all of the new or more casual traders, which would kill the market. You can't have a functional ecosystem with only sharks in the water.

4 people like this

The 10% on profits is fine, there just needs to be a cap on the withdraw fee.


5 people like this

I actually think PredictIt's basic fee structure could be made to work if it were based on NET profits, in a market group rather than gross profits on and individual contract. Monthly fixed fees tend to keep away smaller traders (who might eventually become bigger traders), and trading fees create there own distortions, particularly for extreme prices (below 5, above 95). The latter can be made to work pretty well by lowering the trading fees at the extreme. In any case, the primary source of distortion for trading fees is based on use of gross fees on individuals markets. 


Though I think gk123 is overstating the impact of the fee structure in the GOP presidential nomination market (I think the main distorter there is the lack of margin-linking, requiring large capital outlay for shorting multiple contracts). Nonetheless, the fee structure does create significant distortions. I think these are easier to see in the Party Prez market. Here, the bids on the GOP and DEM yes contracts regularly sum to around 101 or 102 (with the asks summing as high as 108), in addition to the Other contract which has typically had a 4-7 spread.  Margin-linking is still playing a roll here, but even if the market was linked, the return on across the board are limiting by the role of fees on gross profits. For simplicity of the math, assuming you can sell D at 60, R at 40, and O at 5. Without fees, you expect a 5 cent profit for shorting across the board. With fees, you pay 4.5 cents if the Dems win, 6.5 cents of the Rs win, and 10 cents if Other wins, for a net profit of 0.5 cents, -1.5 cents, and -5 cents, respectively. The market has to overbid by about 10%, before it is really worth it to short across the board. If fees were based on net profits in a market, however, there would be no disincentive to sell across the board until the bid total is brought down to 100.  Note that this is true even if the fee percentage is raised substantially to offset the revenue lost due canceling out phantom losses (basically "losses" made in fully hedged positions). I would rather see a fee of 25-30% on net profits than the current 10% on gross profits. 


6 people like this

I couldn't disagree more with freelancepope. 


The 5% withdrawal fee is needed to offset the credit card processing fees the PredictIt incurs when we make deposits. It provides an appropriate incentive for traders to move money in and out of the site more than necessary and does not distort prices or substantially deter new traders. If withdrawal fees were capped, that could lead to abuses where people use PredictIt deposits just for the credit card rewards, making PredictIt (donors, other traders) bear the costs of processing the CC transaction. I really think the only problem with the current fee structure is net vs. gross issue (though I would also prefer fees to be based on average cost rather than FIFO) and that fixing that in a revenue-neutral way would be a huge improvement, second in importance only to rapid implementation of margin-linking.


8 people like this

Appreciate hearing others' thoughts.  A few responses from my perspective:


hvpark:  I don't think it's either/or.  It is fine to continue to allow current fee structure to stand as an option - I agree for many users, paying a subscription wouldn't make sense.  That said, allowing heavier users the option of opting in to a subscription would significantly improve the accuracy of the market, as I pointed out in the original post.  


Freelancepope:  capping the total withdrawal fee has a similar impact to a subscription.  I'd be fine with it as an alternative, if that is somehow easier than setting up a subscription. 


gabeweil:  Credit card processing fees vary, and I agree that it is clear this fee is a partial offset.  That said, if the whole 5% were going to credit card processing, we have found our problem right there.  It would imply nearly all the fees the site takes in are going to 3rd parties.  They could save everyone a boatload by offering to waive/reduce the fee if you link a bank acct or use paypal instead.  In reality, I think a meaningful % of that 5% must be going to support site infrastructure / running the project, and thus it could be assessed in a different manner so that it doesn't distort the market.  


I do agree with your net fee construct and margin-linking being two key issues that would improve the market.  That said, I would point out that a 25% net profit fee combined with a 5% withdrawal fee will still result in gross spreads in most markets being near double-digits before there is an opportunity to create any profits for users.  If no alternative fee model is introduced, the overall market odds will continue to suffer.  


Not to beat a dead horse, but does anyone here really think there is a 94% chance Rubio, Cruz, or Trump is the republican nominee?  (46, 24, 24 right now).  What is the message we should draw from those odds?  I think the only rational thing to believe is that they are surely far too high, and probably mostly useless when trying to draw meaningful conclusions from the data.  Ben Carson is currently trading at an 8% chance of winning the nomination and also an 8% chance of winning the presidency.  Do you think that tells us something about his chances of winning and/or launching a 3rd party bid, or do you think it tells us that fees prevent anyone from taking the "No" side of the bet profitably below about ~8%?  Assuming the research purpose of this site is to aim for a market that produces accurate (or at least useful) results, I don't think this is a small issue.  


Here is the New York Times weighing in on the site's odds: http://www.nytimes.com/2015/10/24/upshot/betting-markets-call-marco-rubio-front-runner-in-gop.html?_r=0


"Traders at Betfair, which is the world’s largest betting exchange, but which doesn’t take bets from Americans, rate Mr. Bush a 20 percent chance to win the nomination, while Mr. Rubio is given a 29 percent chance. Over at PredictIt, which is a small-scale experimental prediction market popular among American hobbyists, traders have moved more decisively, giving Mr. Bush a 24 percent chance, compared with Mr. Rubio’s 40 percent. (A peculiar inefficiency in that market tends to overrate the chances of all of the candidates, so both of these numbers are surely too high.)"


Indeed.


2 people like this
I disagree entriely with the new ideas proposed for the fees. Predictit will lose users if the fee structure was to be per contract or a monthly subscription. I think that the reason many people are trading on predictit and joining is because of the 10% on profits. People don't mind that much but a fee structure would limit trading. The 5% withdrawal fee on money is also a little too much and the tax consequences are it are also vague, so removing that or decreasing it would be perfect in my opinion.

gk123:


How would fees on net profits distort the market? Give me an example of a trade you would make absent a fee based on profits, that you would be deterred from making by a fee on net profits. I totally agree with your points about how distorted and overbid the market is, but with a fee on net profits and margin linking, you would have every incentive to short and push down those prices to a bid total of 100, since you would lock in profits instantly (you can have negative margin positions with margin-linking, as I currently do in the SCOTUS retirement market) and there would be no risk that phantom profits would result in you incurring fees in excess of your real, net profits.


You're probably right that 5% is a little more than needed to cover CC processing fees, but not much more. I think they typically about. 2.5%. There's also probably some internal resource costs in processing the deposits and withdrawals. So, they might be able to cut the withdrawal fee a little bit, but I don't think this should be the focus of concern. If they could processing through bank accounts with a lower fee, I agree that makes sense.




Gabeweil:


First, let me say that margin-linking and 5% fee on withdrawals are bigger factors causing inefficiency vs. the net profits fee (assuming it was changed to be truly net across contracts vs. how it works now).


That said, all fees (or taxes) on activity serve to reduce activity.  This is one of the most fundamental building blocks of economics.  I provided a real market example of how the current fee structure does this, including the 10% net profits fee.  Let's assume you fixed everything else, there was no withdrawal fee, and the only fee in the market was the 25% fee on net profits you proposed.


Now, how much $ do you need to make in order to entice you to make a trade on this market?  Let's say there are mediocre alternative investment options and you're willing to accept an 8%/yr return and so is everyone else in the market.  To give you a specific trade, let's use current SCOTUS market, where I believe the 9 contracts currently add up to 109% on the offer side (i.e. you can buy no on them all and lock in at 9% spread).  Imagine you could guarantee someone leaves the court within a year.  Would you trade?  You would in a world where you are paying a fixed subscription to use and support the site, and you do not in a world where you have to pay a 25% tax on your profits (which would reduce your net return below your hurdle rate).  Thus, in a subscription world, we should expect spreads to settle at ~8% annualized, and in a net profits fee world, you need >10% spreads to make a trade that works.


The point here is not that we need to eliminate all fees.  To be honest, if it were only the 10% net profits fee, I would never have started this thread.  The point is, philosophically, levying fees based on user activity on the site distorts the market, and it creates odds that do not reflect reality, but rather reflect the fee structure and incentives being provided (causing news organizations like NYTimes to dismiss them as totally unrealistic).  I don't really know what the goal of this site is.  If it is to attempt to create accurate odds, or measure whether a betting market can accurately predict events, then allowing users to opt in to an alternate model of payment that is not activity-based (such as a subscription) would certainly help.  If it is to make money or it has some other purpose, then maybe it doesn't matter.







GK123:


First off, let''s clarify what I think we agree. 


1. Moving in a revenue-neutral way (25% is just a number I through out, I suspect it would actually be lower) to net fees would be an improvement, and less distortionary than the current structure. In your SCOTUS example, a 10% fee on gross profits would basically wipe out all gains if the market were at 109% and could even result in losses for an across-the-board shorter if the a long-shot (below 10) result occurs.  


2. Some fee on withdrawals is necessary to cover CC processing fees, at least until they offer an alternative deposit option (with fee waiver / reduction).


3. While my proposed fee structure avoids turning profits into losses, it does reduce expected rates of return such that some trades or strategies will offer less attractive risk/return options than alternative investments, even when the market prices are clearly wrong (e.g., the market is overbid).


Let me know if you disagree with any of that.


I don't really have a problem with PredictIt offering a monthly subscription as option, but I do see some potential issues.


1. It would have to be optional to avoid deterring new traders from joining. At a minimum this would increase complexity and administrative burdens for PredictIt.


2. To be revenue neutral, the fee would have to be set fairly. This is because the most active and profitable traders are the people most likely to take advantage of the option. The fee would have to replace the significant revenue those traders generate. 


3. The higher the fee is set, the fewer traders will join, self-selecting for those with the highest monthly profits. This could require repeatedly raising the fee to make up for revenue shortfalls, possibly resulting in a death spiral. Basically this is an adverse selection problem. I think part of the appeal of a subscription model is that big traders think it will save them money. But if it's going to be revenue-neutral, this really can't be true on average unless there are a lot of small-time suckers willing to pay the fee every month even though they would be much better off just paying the fees on profits.


4. Even if adverse selection doesn't render the monthly option non-viable (maybe they scale it with account balance or trading volume somehow), it is still a cost that would be factored in when comparing alternative investments. If the fee is scaled up somehow to avoid adverse selection, you're going to get back into the problem you raise that the fees lower expected rates of return. There's no really way to get around this problem.


So, while I concede that some distortion would remain under a net profits fee structure, I think it would represent a very substantial improvement over the current structure, without raising the multitude of complications associated with a subscription model. Most of all, I don't want the unwieldy nature and potential pitfall of a subscription model to scare PredictIt off from a more modest change in fee structure that would solve most of the problem.


Finally, I just don't think the 5% withdrawal fee is a big problem. It's not clear to me why it's any more distortionary than any other way of raising revenue. I think both the withdrawal fee on the fee on net profits seem so bad to you because you are comparing them to some ideal fee structure that avoids all distortion. But you get that by assuming away all the problems with the subscription model. If it's going to raise the same amount of revenue (probably needs to do a little more to cover additional administrative burden), it's going to have reduce average rates of return. The big difference (if you could get everyone to sign up for it), is the it would burden less smaller and less successful traders proportionally more, but these are precisely the people who would not agree to pay the subscription. So you either let them keep the current fee structure or lose them entirely.




1 person likes this

Gk123:

I sent a brief feedback message to them before coming to the forums. I'm very glad to see you already made all the points much more coherently than I could have.

My tl;dr point: no withdrawal fees would attract more (presumably smarter) money to the website. The increase in gross trading would yield a higher gross amount being subject to fees on profits. Smaller percentage of a higher number, yaddah yaddah.

The assumption is an increase in users and activity would not cause an increase in costs, and more narrow (I'd say more reasonable) fees would be beneficial for both the accuracy and profitability of the markets.


1 person likes this

One related issue I want to raise is that the current architecture for putting in orders discourages market-making. If you have a net "yes" or "no" position on a contract, you have to fully close out that position before you can put in orders to take a net position in the other direction. For example, if I have hold 30 yes shares on Marco Rubio to win the GOP nomination, I can't put in an order to sell more than 30 yeses. I have to first sell the 30 to close of my old position, then put in a new order to "buy no". This is fine when you have a big net position one way or the other and just adjust the size of your position in response to price fluctations, and it's fine generally in tight markets where all you want to do is make a bet on one side. But when there's a large bid-ask spread and you want to act as a market maker by putting up both buy and sell orders inside that spread (providing liquidity to other traders in the process and making a small easy profit at the risk of the market moving sharply and leaving you exposed), you are hampered from doing that. Similarly, if you want to put in big orders deep in the order book in case there is a big temporary price swing, as happens from time to time, the current structure often prevents that. This seems like is should be a pretty easy fix, and costless once implemented.  This is quite on point with the fee conversation, but I think it is causing similar issues in terms of impairing the efficiency of the market.

gabewell:

I think the basic reason that the 5% withdrawal fee is significantly worse (for market efficiency) than the 10% fee on profit because it sets a lower bound on the rate of return needed to make arbitrage profitable. 

As an example, suppose that the price of Hillary's No shares in DNOM plus the price of Bernie's No shares in DNOM is 0.9. This compound purchase gives a guaranteed return of 1, so the pre-fee rate of return of this trade is 1/0.9 = 1.11. If you spend $90 and buy 100 No shares for each candidate, then your pre-fee revenue would be $100, so your profit is $100-$90 = $10. The 10% profit fee brings this profit down to $9, which means you are still earning a rate of return of 99/90 = 1.1. 

By contrast, if you wanted to withdraw this $99 from your account, you would only get $94.05, so this fee brings the rate of return down to 1.045. You see how significant this difference is?


4 people like this

I'm new to all this, but would a (credit-card processing) fee on DEPOSITING funds take care of the market distortions? I suppose after that, the 10% on profits is Ok with me. Then there could be a nominal flat fee for issuing checks and postage. It seems weird to me that there's a percentage to withdrawal that's presumably applied to my deposit. Applying a deposit fee should yield the same profit for PredictIt. It would just charge everyone on the way in instead of only the winners on the way out.


2 people like this
Login to post a comment