Now that PDS has successfully launched, I want to direct some attention to a brand new project, one that has the potential to benefit a large number of traders. It is called The Slippage Project. This project will rely on crowdsourcing to quantify the average amount of slippage¹ one incurs when using market orders on the open. Before I get into the specifics of how this project will be implemented, some background information is necessary.
What Is the Opening Auction or Opening Cross?
When one places a market order to be executed at the next open (market-on-open or MOO), he or she is agreeing to participate in the opening auction or opening cross. In simple terms, during the opening auction, the stock exchange runs an algorithm (at least on the Nasdaq) and attempts to match buy orders with sell orders. If there is an imbalance such as more buy orders than sell orders, the stock will gap up as there is more demand than supply.
Link to NYSE Arca Equities Opening Auction
Link to How Nasdaq’s Opening and Closing Cross Works (Highly recommended reading).
When we agree to participate in the opening auction or cross, we accept that we will get the opening price. However, and this is very important, there can actually be different opening prices for each security!
The Official Exchange Open vs. The Consolidated Open
When a stock opens for trading at 9:30 a.m., the cross that takes place on the major exchange, e.g., the NYSE or Nasdaq, is recorded as the Official Open price. While the Nasdaq is very quick at crossing trades, the NYSE is very slow, primarily because the Nasdaq is an electronic exchange while the NYSE is still suffering from the remnants of the specialist system. If a trade goes through an ECN before the opening cross is finished (this happens often on the NYSE, remember it is slow) the ECN trade is recorded as the consolidated open. For example, while the NYSE is busy counting on their fingers and toes, trying to match orders for ABC stock, an ECN may match a trade at 9:30:01 a.m. (making the consolidated open). ABC stock may not officially be opened by the NYSE until 9:31:00 a.m, and when it does open, the official open price is recorded.
So to recap, we have an official open price, and a consolidated open price. The problem is that on the NYSE, the two prices are usually different!
To make matters worse, most data providers, including Yahoo, Google, Tradestation, Interactive Brokers, etc. publish the consolidated open price, and NOT the official open. And here is the critical point: Most retail traders could never actually get the consolidated open price, only the official open price. Yet the consolidated open is what is commonly reported.
The Problem, and a Solution
The problem is that we are presented with an opening price that is unattainable.
When we backtest strategies that trade the open or the close ( more on trading the close later), we are assuming that we could have purchased a stock at a price that would not have been obtainable in the real world.
For strategy traders, if on average we are filled at a price higher than the consolidated open, it will mean that our strategy may not perform as well as it did during backtesting. Similarly, if we consistently get filled beneath the consolidated open, it will mean that our strategy may actually exceed the performance results derived from backtesting.
The solution is to record hundreds if not thousands of opening buys and sells, and calculate the differences between the official opens and the consolidated opens. After recording enough opening prices, we can determine a realistic figure to use during backtesting.
My friend BHH and I started recording our opening trades, some time ago (credit must go to him for the idea). When we stopped logging trades, we were actually seeing a positive difference, meaning that on average we were getting filled at prices lower than the published opens.
Tito over at Hack the Market wrote about a similar project where he logged roughly 850 trades, and saw a price improvement of .04% on the trades where he was not filled at the published open.
For us strategy traders that trade the open or the close, this information is invaluable. We may be seeing positive price improvements that will negate the cost of commissions, or not. The alternative is that we should be building higher costs into our systems to account for negative slippage.
Implementation
I do not want any exploding heads, so implementation of the project will be discussed in the next post. Stay tuned! This is a project that will benefit many of us retail traders, but in order for it to be successful, we will all have to participate!
¹ This is not slippage in the traditional sense of the term Rather, we are talking about the difference between the official and consolidated opening prices.
> The solution is to record hundreds if not thousands of opening buys and sells, and calculate the differences between the official opens and the consolidated opens. After recording enough opening prices, we can determine a realistic figure to use during backtesting.
This is a sort of solution, but a more correct (albeit pricey) solution is to buy better data so you have the correct opening prices by venue. I like the crowd-sourcing idea, but there is a correct/actual opening price out there, so finding and using it would seem to be the most sensible and direct way of addressing this issue… My intent in looking at this originally was to get an idea of the scale of the issue. On average it was small but I felt it was significant enough to warrant getting better data and that’s ultimately what I did.
Good luck with your efforts!
Thanks Tito. I agree with you completely, but for the average retail systems trader, even a well-capitalized one, the price of the data is very prohibitive. I am curious where you decided to get your data, if you don’t mind sharing.
I think the exercise will be very beneficial if we can get a large number of traders and a large sample size.
I get it from the nyse directly – the taq. The cost isn’t bad if it helps answer questions like “what’s the typical volume on the open” or “what was the real price at the open” and oh so many more. Which it does. The *real* cost, though, is that building out a h/w infrastructure and the s/w that can intelligently access it to answer those questions is expensive and time consuming. You’re getting over 50G/day so I think it qualifies as a “big data” problem.
Your proposed project may be beneficial, but I’d probably just accept that the published price is close and that the volume on the open can’t be expected to be more than a fraction of 1% of the overall average daily volume. Size accordingly and then track how you do in reality against what the printed open price would’ve indicated. My little experiment suggests that it’s not likely a grave source of error, but my later look at the real data and subsequent experience suggests that the real gotcha for moos is volume. If you place a moo with volume that’s a significant fraction of what you might expect at the open you’re likely to pay a penalty and would likely be better served trading near instead of at the open.
Hi, Wood:
I mentioned the problem of opening prices a few posts ago because of observations about another subscription service that developed a following and started having impact on prices. I think the question is: what data can be used that lets us test systems and expect that the implementation will match the test. My observation is that unless a stock has an active pre-open market, the liquidity at opening is very sketchy, taking seconds to minutes to get to a normal spread. Once a price has been moved by an imbalance of orders, that becomes an anchor for subsequent trades — they won’t soon move back to the opening price you wanted or expected based on testing.
Unless it’s SPY or AAPL or such, I just assume that opening prices are not implementable, so I don’t even test with opening prices on normal stocks.
Top, once I can get a large sample (assuming I can) then some observations can be made.
I think there are other considerations such as is the trade a mean-reversion setup or trendfollowing? Is any effect exaggerated on companies with smaller capitalization? Are some brokers giving better fills than others? What role does routing play? Odd lots?
I hope to be able get enough data to make some general observations about all these issues. And then that will drive future testing and decision-making, especially in regards to limit orders rather than markets.
PS — thanks for the research on opening auctions!