other equivalent time periods.
Market-makers have been arguing about this for a long time. One
group said it was smart to sell options to “collect theta.” The other
side said that idea was stupid because an edge could only exist
when one side had an information disadvantage. Everyone had a
calendar, so why would there be an edge purely due to the passing
of time? The market would price the risk of holding options over
the weekend correctly.
The “smarter” traders were in the second group. They were wrong.
The options market does not correctly price weekend decay. It is
profitable to sell options over the weekend.
Christopher Jones and Joshua Shemesh studied this issue (2017).
They looked at the returns of long option portfolios on US equities
from 1996 to 2007 and found the average return over the weekend
was negative (0.62%) while the returns for all other days were
slightly positive (0.18% a day).
Having established that weekend returns are significantly lower
than those of other days, the authors went on to study other
holidays, including long weekends. Their hypothesis was the effect
was directly related to non-trading, which would imply lower
returns would also be associated with other holidays, and the
effect would be stronger over long weekends. This all seems to be
true: returns on equity options are negative whenever the market
is closed. It seems the effect exists because market-makers are not
correctly adjusting the implied volatilities on Fridays to account
for the upcoming weekend.
This effect is significant. There is no general edge in selling many
stock options (unlike index options, when being short is normally
the way to lean), so this is a totally new effect, rather than a matter
of just timing an entry. Also, the effect was consistent across years
and was robust with respect to exactly how the portfolios were
constructed.
It seems likely that the same effect exists with index, bond, and
commodity options, but this is untested.
98
Trading Strategy
On a Friday, sell the options that expire the next Monday. In
general, time any short-volatility strategies to include as many
non-trading periods as possible.
Volatility of Volatility Risk Premia
Options on products with high volatility of volatility tend to be
overpriced. This is true both in the cross-section—options on
stocks with high volatility of volatility are overpriced relative to
options on stocks with low volatility of volatility—and in the time
series—the VIX tends to decline (rise) after very high (low) values
of the VVIX index.
The first empirical study of this effect was done by Ruan (2017).
Using data on US equity options from 1996 to 2016, he found that
ranking stocks by the volatility of their implied ATM volatility
showed that there was a strong and consistent negative
relationship between delta-neutral long option positions and
volatility of volatility.
A similar study was independently carried out by Cao et al. (2018),
who also studied US equity options. Again, using data from 1996
to 2016, they found that the delta-hedged returns of long option
positions decreased in uncertainty of volatility. This was true
whether they used implied volatility, time series volatility from
daily returns (specifically EGARCH), or high-frequency volatility.
Their results were robust with respect to idiosyncratic volatility,
jumps, term structure, the implied–realized spread, liquidity,
analyst coverage, and the Fama-French factors. They also showed
that the effect was largely driven by volatility of positive volatility
moves, and that volatility of negative volatility moves had a
negligible effect.
These studies leave little room to interpret this effect as anything
other than a separate volatility of volatility premium. Ruan (2017)
just states, “Investors indeed dislike uncertainty about volatility of
individual stocks, so that they are willing to pay a high premium to
hold options with high VOV [sic],” with no supporting argument.
Cao et al. (2018) speculates that market-makers were charging a
higher premium for options with high uncertainty of volatility,
because those were more difficult to hedge. This might be a partial
99
reason, but it doesn't take into account the time-series result that shows high volatility of volatility predicts a fall in subsequent
implied volatility. This effect is independent of hedging issues.
The relationship between high VVIX (the model-free implied
volatility derived from VIX options) and subsequent lower VIX
levels is very strong. Using VVIX data from 2007 through 2018, I
calculated the rolling 1-year 90th percentile of VVIX. Going
forward, if the VVIX crossed above this level, I “sold” the VIX and
“held” until VVIX reached its rolling 1-year median. This produced
31 hypothetical trades. The total “profit” was 108 points. Twenty-
seven trades were winners. “Buying” the 10th percentile was also
“profitable,” making 62 points over 35 trades, 26 of which were
winners. Clearly this particular idea cannot be implemented
because the VIX is not a traded product, but I've included it to
show that extreme VVIX is a strong predictor of the VIX (however,
if we traded VIX futures, the idea is still profitable). No
optimization was attempted. The idea also works if we use
different look-back periods or moving averages instead of
medians.
This effect has been studied (far more rigorously) by others.
Huang et al. (2018) showed that volatility of volatility significantly
and negatively predicts delta-hedged long option payoffs. Park
(2015) showed that high levels of VVIX raised the prices of S&P
500 puts and VIX calls and lowered their subsequent returns over
the next three to four weeks (a similar time period to the average
holding period in my simple test). He speculates that the effect is
caused by either “risk premiums for a time-varying crash risk
factor or uncertainty premiums for a time-varying uncertain belief
in volatility.” Both of these are plausible but at this point there is
no independent evidence for these causes.
Trading Strategy
When VVIX reaches extremely high (low) levels either sell (buy)
VIX futures or sell (buy), and dynamically hedge, S&P 500
straddles.
Confidence Level One
The confidence-level-three strategies should form the core of a
trading operation. But the ideas that I think are true but only give
100
a confidence rating of one are also important. Trades based on
market inefficiencies will be most profitable when the evidence for
them is still underwhelming. Many inefficiencies will not survive
long enough to reach my level three. So, although I wouldn't
allocate a great deal of my portfolio to these ideas, they can still be
very profitable.
They also offer a way to deal with the desire to gam
ble. Many
traders overtrade and need to always be involved in the market.
Instead of denying this tendency, it is better to accept it and learn
to accommodate this need by tinkering with small trades that still
have expected edge. Level-one trades are perfect for this. This is
like the idea of a “cheat meal” when dieting. Instead of trying to
religiously stick to a diet it is better to accept that temptation
exists and schedule regular times when you can eat garbage.
Dieting increases cravings (Massey and Hill, 2012). There is solid
psychological research that shows that dieters who include cheats
do better than those who don't (do Vale et al., 2016). I expect that
active traders are tempted to over-trade and that cheating helps
them as well.
Remember that cheats, whether in dieting or in trading, need to be
kept small. If every meal is a cheat meal, you aren't on a diet. You
will just get fat. And if every trade is a speculative one, you aren't a
disciplined trader. You will just lose money.
Earnings-Induced Reversals
Earnings-induced reversals are the tendency of stocks that have
drifted a lot before their earnings announcement to reverse the
pre-announcement drift when the news breaks. This effect was
first studied by So and Wang (2014). Using US equity data from
1996 through 2011, they created a trading strategy that shorted
stocks with high market–adjusted returns in the period from four
days before earnings through two days before earnings and went
long those stocks with the worst pre-earnings market adjusted
returns. (This seemingly odd time period was so that they could
trade on the close of the day before earnings without using the
trade price when choosing the portfolio. Obviously, a trader using
intra-day data could use a different time period without
“cheating.”) Liquidating the portfolio on the close after earnings
they found this portfolio made 145 bps compared to the 22 bps
101
earned by a similarly constructed portfolio during non-earnings periods.
A similar study was done by Jansen and Nikiforov (2016). Simply
fading stocks with large percentage moves in the week before
earnings would have averaged 1.3% over 2-day periods.
Jansen and Nikiforov (2016) speculate that the effect is due to
investor overreaction in the pre-earnings period. Individual
investors fear that they are missing information and trade in the
direction of price changes, fueling the trend. After the
announcement, the fear of being ignorant of information goes
away and the pre-earning return is seen as excessive. This might
be true. A similar effect is seen in sports gambling, when “steam
chasers” bet on teams that have shortening odds on the suspicion
that smart money is driving the price changes. But much more
would need to be done before I am confident that this is an
inefficiency. Currently the statistics are inarguable, but the
reasons for them are close to a mystery.
Trading Strategy
I'm more confident in the collapse of implied volatility when
earnings are released than I am of this reversal effect. So, when I
trade both of these effects together I sell a straddle but shade the
delta if I want to also bet on the reversal.
Pre-Earnings Announcement Drift
Pre-earnings announcement drift is the tendency of stocks to
move in the direction of any earnings-related abnormal returns
experienced by stocks in the same industry that reported earlier.
This effect was first studied by Ramnath (2002), who investigated
how information from the very first earnings announcer within
each industry (the 30 industries identified by Fama and French,
1997) affects the prices of later announcers. He found that the
earnings information for the earliest announcing firm within an
industry predicts both the earnings surprise and the returns of
other firms within the industry.
This effect was later confirmed by Easton et al. (2010), who used
not just the first reporter in each industry but also the effect of all
the earlier announcing peers.
102
The drift begins as the results from the early announcers are
reported and continues up until the later announcing stock
releases its earnings. The effect is above the industry beta, which
measures the normal relationship between returns. If earlier
reporting stocks all rally, we would expect later reporting stocks to
also rally just due to industry exposure. Pre-announcement drift is
a separate effect.
Pre-earnings anomalies have not been studied nearly as much as
post-earnings anomalies, so the evidence is comparatively weak,
and it is not clear what causes the drift. As with PEAD, the pre-
earnings move is plausibly due to underreaction to new
information: here the earnings of the related companies. It could
be that investor overconfidence causes them to be anchored to the
pre-earnings price and incorporate the new information only
slowly. A lot more study would be needed before we could be
confident in this explanation. But there is no obvious risk factor
that could explain the drift, so I would say, tentatively, that this is
a market inefficiency.
Trading Strategy
As we also expect implied volatility to increase in the time leading
to the earnings release, any long volatility directional strategy
would be appropriate. For example, if we expect a rally, we could
buy a call or call spread. My preference is for a 50 delta/20 delta,
1-month call spread. But tastes vary.
Conclusion
The idea that trading edges disappear as soon as they become
public is an oversimplification. Markets vary in their ability to
absorb new volume. A published edge will persist longer in the
S&P 500 than in soybeans. Further, crowding affects different
strategies in different ways. And risk premia will survive longer
than inefficiencies.
But unless noted, the edges listed in this chapter have been robust
until now. It is quite possible that their size will diminish and even
disappear but we have a fairly basic choice: go with the effect that
has worked in the past and hope it continues or choose to do the
thing that would have lost money in the past. Your choice.
103
Summary
It is worthwhile to search SSRN periodically to find new
trading ideas.
Many volatility trading edges involve selling options in
situations of uncertainty. This can be viewed as an extra,
situational variance premium.
Because of the variance premium, long-volatility strategies are
unlikely to have as much edge as those that involve selling
options.
104
CHAPTER 6
Volatility Positions
One of the things that make options great is that there are many
ways to express an opinion. But this is also one of the things that
make options tricky. Just because there are many ways to express
an opinion doesn't mean they will a
ll be equally good. The
differences are not trivial. Some will be a lot worse than others.
In this section we will compare some option positions that are
primarily used to express views on volatility. We will look at the
possible distribution of returns by using both GBM returns and
historical data. We will also look at the effects of the underlying
having a drift. This will generally be done from the perspective of a
volatility seller, but the case of long volatility is a trivial extension.
All of the simulations will assume that we initiate the position and
then leave it alone until expiration. In reality, we will usually have
opportunities to trade out of the position before then. But it is
important to understand the terminal distribution of the P/L for
several reasons:
Even an adjusted (or hedged) position is instantaneously
subject to the same issues as one that won't be adjusted in the
future.
Very short-dated options (depending on the market liquidity
this could be weekly, daily, or hourly) can't meaningfully be
adjusted.
The actual adjustment procedure will be different for different
traders so will be impossible to simulate.
Aside: Adjustment and Position “Repair”
“Repair” is a dangerous misnomer. First, in any other situation to
repair something is to return it to its previous condition. But in
the trading world it is usually taken to mean turning a losing trade
into a winning trade. This is a falsely reassuring idea, but it can't
be done. The loss is already in your account. That money is gone.
Forget about the original trade and ask yourself, “Given what I
105
now know, what position do I want?” Then put that position on.
This is completely independent of the original trade. This should
also be done when examining winning positions. Their profits are
also in the past. Do you like the position now? If not, do
something else.
You should adjust a position when it no longer matches your
forecast or opinion. This is true whether it has previously made
money or lost money.
Straddles and Strangles
The two most basic ways to short volatility are to sell either a
Positional Option Trading (Wiley Trading) Page 12
