Why do option MMs misprice options intraday?!

[goldorak]

Details. The truth is if you integrate even minimum market impact on the open and close auction due to your own participation in the process, results are not all that nice anymore. Btw if someone knows of any study, academic or not on market impact during open and close auction, I am a taker... Intraday market impact of orders and meta-orders, there is plenty. When it comes to open and close auctions, the emptiness of the academic/practitioner literature on that particular topic is upsetting.



On top of that academics often use data from more than 10-15 years ago. Sorry to have to let them know that open price did not mean the same as today. It is not the open auction price that you see in most cases, even for a number of stocks in recent data.

[EspressoLover]

In practice, it's pretty common to use 10:00 AM continuous trading prices. Liquidity is atrocious at opening auctions. Most of the common overnight effects still hold up if you measure this way. Closing auction's pretty fair though. Liquidity's thick, adversity's low and market impact tends to be smaller than continuous trading.



Not many studies on auction impact (best I know of is here). But in general, estimating market impact at auctions is easy compared to continuous trading. If you have the order data, you can easily derive the change in auction price from adding an additional order. The only empirical question is the market response to the imbalance broadcast. But that's straightforward to regress.

[goldorak]

Thank you for the paper. I knew about it. Not really exciting and a bit outdated. I guess you will agree with me. Nothing to be compared to studies from Bouchaud & co for example.



> If you have the order data,



which you unfortunately don't have access to. As far as I know the only data you can get about the auction, apart from matched volume and auction price, are imbalances. Or did I miss something?



> The only empirical question is the market response to the imbalance broadcast. But that's straightforward to regress.



The 1bio$ question. Anyway, try to put 1mio$ orders at the open on stocks with apparently juicy overnight returns with decent model of market impact and low even extra low broker/exchange/regulator fees. The nice academic results do not look all that good anymore.

[murfury]

Thank you all for constructive comments and for your interest!



I reply to common points here, and then to specifict points one-by-one below.

1. The overnight strategy is not profitable after costs (spreads are >6% of opt price) unless you execute your trades really good (I have a separate paper about how to do this). But the day-night effect can be used to reduce transaction costs. Last section of the paper elaborates on the trading strategy implications.

2. Like many of you, we thought that the inablity to hedge overnight jump risk drives the option returns. Surprisingly, we didn't find much evidence of this. Overnight reutrns do not depend on measures of jump and tail risk. Overnight and intraday returns are equally negative after controlling for the volatility bias. Maybe you can suggest some futher tests for the overnight risk we can run?

3. Unfortunately, day-night anomalies in options and equity market are unrelated. First, no day-night effect in S&P500 futures in our sample period. Second, we delta-hedge option returns and include market return in robustness regressions - we don't want to reinvent equity market anomalies. Third, the options effect is just so much bigger. Finally, we find a clear explanation of the options day-night effect but no idea how to explain the equity patterns. Also, a recent paper argues that the equity day-night stuff is quite unstable for international markets (data-mining?).



We appreciate any suggestions on how we can further test for these three points.

[murfury]

ronin: you’re right - making money using the overnight strategy is tough (ba spreads are typically > 6% of option price). My point was that this strategy can definitely help reduce execution costs.



sigma: you’re right, I compute average returns, but your suggestion to condition on news is interesting. I use option quote midpoints, we do have data on actual option trades but decide not to compute open/close prices with it because of microstructure and other problems it creates.



EL: You second comment is interesting. I agree that equity MMs are often flat overnight, but OMMs have to carry inventory for long periods, and thus OMMs are keen to predict mid-term IV changes. Your idea that the day-night effect is driven by client order flow was one of the first we tested. We didn’t find much of the relationship between the two. Maybe we can test this hypothesis better?

Also, interesting EAD hypothesis, you have in mind option returns or stock returns? We focus on index options, and individual EADs usually have small effect on the index, but we can certainly test it.



radikal: not sure I fully understand the slang in your terminology. Overnight period might be more risky because MMs carry tail risk, but then it doesn’t explain positive intraday returns. Also, we didn’t find much variation in overnight returns then days are sorted on tail risk measures. I’m open to how else can we test the tail risk idea.

[murfury]

Sigma: thank you for sharing these statistics. They’re certainly very informative.

[EspressoLover]

@murfury



I finally got a chance to read the entire paper. My overall opinion: Two thumbs up! I'm impressed about how thoroughly you guys investigated the various angles. Couple quick points that stick out to me:



1) FTA: "A delta-hedged option portfolio has zero delta and thus zero beta". This statement is not true ipso facto. Because of the leverage effect, equities IVs have negative correlation to the underlying. Your paper actually even directly refutes this in Table 11. The delta hedged index-option portfolio still exhibits statistically significantly negative beta exposure. Not to be nit-picky, since you still do find a significant anomaly, even with the beta adjustment. But generally be careful about making assumptions about delta-neutrality implying beta-neutrality.



The other consideration may be to evaluate the index-options using 16:00 mid-quotes as an implied close. The results probably are not substantially different, but it could have an impact. This sticks out, because you're not finding an overnight equity premium in the futures, whereas others are in the ETFs. The difference in closing time may be the issue. Also a non-negligible proportion of earnings announcements occur between 16:00-16:15.



2) W.r.t. to client flows, I may be missing in the paper where you tested. But as far as I can tell, your measures of order flow suggest that it *does* drives the effect. In Table 7, the pre-close 5th period seems to have the most long-option order flows. In 3 of the 4 categories (besides equity puts), order flow imbalance is strongest towards long-vega during this period. Similarly the beginning of the day seems to be the most imbalanced towards selling options (i.e. selling vega).



Assuming that OMMs are mostly sitting on the other side of client-initiated trades, this is consistent with a flow driven explanation. "Real" traders seem to bias going long vega near the end of the day. This bids up IVs and options prices at the close. Options become systematically too expensive to buy at the close. Vice versa for the open. Overnight returns become a repeated game of buying high and selling low.



3) W.r.t. to earnings announcement. You may be able to get a rough estimate of its effect by using a proxy variable in your robustness regression. E.g. number of companies in the index announcing earnings that night.

[ronin]

@murfury



I had a slightly closer look now.



My main comment is about your use of mid prices. Mid price, as in (bid+ask)/2, is generally meaningless - even for liquid instruments, and even more so for options.



You would be better off fitting an arb free vol surface that stays inside all the spreads, and looking at what happens to this vol surface.



The alternative interpretation based on your use of mids is that asks are quicker to form (mms sell options, they rarely buy) and that bids only fully form leading up to close. But because you are looking at raw mids, to you it looks like options are cheaper in the morning when it is really that bids don't fully form in the morning.

[murfury]

EL: Thank you for your kind words, we put a lot of effort into the paper!



I totally agree with your statement about delta-hedging. What model is typically used for delta-hedging in practice? I remember Emanuel Derman once said that there is no consensus on whether deltas should be above or below their Black-Sholes values. Is it still the case (no consensus)?



Your points about using 4pm closing time and order flow are also well-taken. Let me just note that theoretically, prices should immediately reflect any future expected order imbalance (e.g., expected selling pressure tomorrow will cause prices to drop today – well in advance of actual order imbalance).



Ronin: We are also concerned about using mid prices (most option papers do), but don’t see a good way around this limitation. We are currently working on re-computing option returns using only ask or only bid prices as you suggested. Vol surface is hard to use for our purposes because it’s normally computed using calendar time, i.e., we need to back out option market’s expectations about tomorrow’s variance and compare it against actual variance we see on that day (I struggle a bit on how to compute these two variables in a comparable way). The fact that we find the day-night effect for all moneyness and maturity categories is reassuring though.



At a big picture level, our results suggest that OMMs seems to use a relatively simple model to forecast vol that probably includes first-order effects (vol clustering, mean-reversion, leverage, earnings ann., M&A) but ignores less obvious stylized facts such as vol seasonality. If this hypothesis is correct, then market under-reaction to these other stylized facts can be perhaps used to generate trading signals.

[radikal]

@murfury --



EL's most recent comment, specifically #2, mostly sums this up:



"2) W.r.t. to client flows, I may be missing in the paper where you tested. But as far as I can tell, your measures of order flow suggest that it *does* drives the effect. In Table 7, the pre-close 5th period seems to have the most long-option order flows. In 3 of the 4 categories (besides equity puts), order flow imbalance is strongest towards long-vega during this period. Similarly the beginning of the day seems to be the most imbalanced towards selling options (i.e. selling vega).



Assuming that OMMs are mostly sitting on the other side of client-initiated trades, this is consistent with a flow driven explanation. "Real" traders seem to bias going long vega near the end of the day. This bids up IVs and options prices at the close. Options become systematically too expensive to buy at the close. Vice versa for the open. Overnight returns become a repeated game of buying high and selling low."



I'm just making the further point that even the OMMs themselves are buying options end of day and selling them the next morning.



With less jargon-y speak, OMM:

- trades reasonably close to flat greeks during day session and to his model is "flat risk" as end of day approaches (by which I mean little to no CORRETED greeks)

- realizes that despite what model says, he may need to buy some additional downside protection because my model doesn't agree with what internal risk says or because my model isn't giving me a terribly margin efficient portfolio to the exchange's calculation

- therefore I buy as many downside puts (or upside teenies) to make everyone but myself happy -- I expect to bleed out some small amount of theta on this but it's generally a small fraction of daily pnl

- in the morning, to reduce theta bleed on hedges I don't want/need to my model, I sell these all back out