Regarding the MOC imbalances, I believe that NYSE publishes them 10 minutes before market close, not 15 minutes, so the linear regression you've shown with "Last 15-min Returns" may have some lookahead bias. Have you tried rerunning the regression with "Last 10-min Returns", and compared the results?
You're correct, and thanks for flagging it! NYSE publishes the imbalance data at 3:50 PM ET, not 3:45. Using "last 15-min returns" (from 3:45) as the dependent variable implies you'd have the imbalance signal 5 minutes before it's actually published — that's lookahead bias.
A few clarifications:
The charts are simulated, not historical. I should have been clearer about this (it's stated in Part 4 of the series but should be repeated more prominently). The regression is illustrative of the documented relationship between MOC imbalance and closing returns, not a backtest of a trading strategy using actual NYSE imbalance data. The published research (e.g., Nasdaq's own analysis) confirms the relationship using correctly timed data.
The timing is more nuanced than "nothing until 3:50." NYSE floor brokers receive indicative imbalance information starting at 2:00 PM, and this information leaks into prices well before 3:50. Before the COVID floor closure in 2020, DMMs were pricing in roughly 40% of auction volume through D-Orders using this early information. So the economic signal starts earlier than the formal publication — but you're right that a retail trader can't act on it until 3:50.
Honestly, IMO if I reran the analysis properly on real data, the correct specification would be last-10-minute returns (3:50 to 4:00) regressed on the 3:50 published imbalance. The R² would likely be lower than what I showed, because 5 minutes of the price adjustment would already be in the data by the time you observe the signal. Nasdaq's own research found that about 80% of the ultimate price move is priced in within 300 milliseconds of the imbalance publication — which means the trade-able edge after observing the signal is tiny for anyone who isn't co-located.
I'll update the chart label to "Last 10-min" and add a note about the timing. Good eye. :)
Two options as a retail trader. Market Chameleon is one. You can track persisting imbalances over multi-day windows and they have some historical data. TBH it is more of an analytical overlay than a raw feed.
The other one is MiM (Market Imbalance Meter) from MrTopStep. It provides intraday snapshots of the auction as it develops, with historical records. Subscription is around $100/month — reasonable if you’re specifically researching closing auction dynamics.
Hey I thought this was a great read, thank you for sharing. I wanted to ask what you thought about the overnight return "anomaly" in terms of pre-berkman, post-berkman(2012), and post-ameritrade(2018). From what I know Berkman 2012 formally introduced this, framing the idea around the thought that retail traders tend to overreact to overnight information, causing price to move disproportionately to actual value, then the price reverses. I recall reading from that paper that this signal was difficult to trade practically as you've pointed out in your "can you trade this" section. TD Ameritrade (now absorbed by Schwab) was the first of it's kind to allow retail traders to put overnight limit orders. I wanted to ask if you had any insight on the affect of Amertrade on overnight trading, did it change the anomaly in any way? Strengthen it? Weaken it? Mostly I wanted to ask whether you decided to split your analysis of this anomaly into 3 periods (pre/post berkman, post ameritrade) like I have been thinking was intuitive.
Hi, Excellent article. Thanks for sharing.
Regarding the MOC imbalances, I believe that NYSE publishes them 10 minutes before market close, not 15 minutes, so the linear regression you've shown with "Last 15-min Returns" may have some lookahead bias. Have you tried rerunning the regression with "Last 10-min Returns", and compared the results?
You're correct, and thanks for flagging it! NYSE publishes the imbalance data at 3:50 PM ET, not 3:45. Using "last 15-min returns" (from 3:45) as the dependent variable implies you'd have the imbalance signal 5 minutes before it's actually published — that's lookahead bias.
A few clarifications:
The charts are simulated, not historical. I should have been clearer about this (it's stated in Part 4 of the series but should be repeated more prominently). The regression is illustrative of the documented relationship between MOC imbalance and closing returns, not a backtest of a trading strategy using actual NYSE imbalance data. The published research (e.g., Nasdaq's own analysis) confirms the relationship using correctly timed data.
The timing is more nuanced than "nothing until 3:50." NYSE floor brokers receive indicative imbalance information starting at 2:00 PM, and this information leaks into prices well before 3:50. Before the COVID floor closure in 2020, DMMs were pricing in roughly 40% of auction volume through D-Orders using this early information. So the economic signal starts earlier than the formal publication — but you're right that a retail trader can't act on it until 3:50.
Honestly, IMO if I reran the analysis properly on real data, the correct specification would be last-10-minute returns (3:50 to 4:00) regressed on the 3:50 published imbalance. The R² would likely be lower than what I showed, because 5 minutes of the price adjustment would already be in the data by the time you observe the signal. Nasdaq's own research found that about 80% of the ultimate price move is priced in within 300 milliseconds of the imbalance publication — which means the trade-able edge after observing the signal is tiny for anyone who isn't co-located.
I'll update the chart label to "Last 10-min" and add a note about the timing. Good eye. :)
Fantastic, thanks so much for the clarification :)
Very interesting. What do you use as a historical data source for the MOC imbalance data?
Two options as a retail trader. Market Chameleon is one. You can track persisting imbalances over multi-day windows and they have some historical data. TBH it is more of an analytical overlay than a raw feed.
The other one is MiM (Market Imbalance Meter) from MrTopStep. It provides intraday snapshots of the auction as it develops, with historical records. Subscription is around $100/month — reasonable if you’re specifically researching closing auction dynamics.
Hey I thought this was a great read, thank you for sharing. I wanted to ask what you thought about the overnight return "anomaly" in terms of pre-berkman, post-berkman(2012), and post-ameritrade(2018). From what I know Berkman 2012 formally introduced this, framing the idea around the thought that retail traders tend to overreact to overnight information, causing price to move disproportionately to actual value, then the price reverses. I recall reading from that paper that this signal was difficult to trade practically as you've pointed out in your "can you trade this" section. TD Ameritrade (now absorbed by Schwab) was the first of it's kind to allow retail traders to put overnight limit orders. I wanted to ask if you had any insight on the affect of Amertrade on overnight trading, did it change the anomaly in any way? Strengthen it? Weaken it? Mostly I wanted to ask whether you decided to split your analysis of this anomaly into 3 periods (pre/post berkman, post ameritrade) like I have been thinking was intuitive.
Do you have a discord?
Nope. I trade and when I find time between trading and family, I post content here. That's about it.