Debunking the Williams Hedge (Part 3)

Today I will continue discussion of the Williams Hedge (WH), which I introduced here.

I mentioned in Part 2 that I would check with OptionNet Explorer (ONE) Support about the IV calculation. He called it a proprietary approximation using option prices “similar but not identical to the VIX style calculation.” That’s all I need to know.

IV % Chg, surprisingly, seems to be a bigger puzzle. ONE Support says this is calculated off the official closing price. As an example, here is a closing screenshot:

Here is the next day at 3:55 PM:

9.78 to 6.39 is a 34.7% decrease yet ONE shows -39%. In case it takes a few minutes for prices to settle to get the official close, I can calculate based on the previous day’s 4:05 PM (9.76) and 4:10 PM (9.79) values. These correspond to IV decreases of 34.5% and 34.7% decreases, respectively. Where does the -39% come from?

I am fine with vendors citing “proprietary methods” in lieu of answering my questions as long as I know what the metric means and that it is consistently calculated. As a basic math fact, percent change is not proprietary and defines one particular correct answer that is not seen above. We still need some clarification.

Speaking of inconsistency, I came back after lunch, powered up the computer, and launched ONE again only to see this displayed for the next day at 3:55 PM:

The 9.32 here is much different than the 6.39 seen above. The percent change is also wrong (should be -5%). I still need clarification.

Let’s get back to talking about the WH itself.

Remember the PCS is long-dated while the long puts are short-dated. When the long puts expire, the PCS will remain uncovered. For example, if the PCS average 90 DTE and the long puts average 48 DTE, then vertical spread risk will remain for six weeks (42 days) after the long puts expire. Strategy guidelines dictate closing the verticals for $0.10 when available. Suppose the spreads may be closed for $0.10 with three weeks to expiration. This leaves us uncovered for 90 – (48 + 21) = 21 days or three weeks. During that time, we will have three uncovered PCS tranches. With 10 contracts, this is a potential gross risk of $300K for 100-point vertical spreads.*

The strategy makes no provisions for PCS that go ITM and potentially expire ITM. Theoretically, this is okay assuming the long puts have exploded in value. I did not see this occur in the backtest, though. Why?

The WH video points out that when the market crashes, horizontal term structure can go from contango to backwardation. Aside from negative delta, this contributes to an explosion in long put premium. What is good for puts already purchased in the campaign is bad for puts still to be purchased as a campaign. In order to continue finding high IV puts for $0.45 or less, we will need to go closer and closer to expiration.

I will continue the discussion next time.

* — This was my width for the backtest. I could have gone 75 points wide, which would have
       forced selling PCS NTM to recoup the extra cost.

Debunking the Williams Hedge (Part 2)

Today I continue discussion of the Williams Hedge (WH) and my backtesting.

Let me address one technicality with regard to SPX IV. When people think about implied volatility for SPX, they usually think of VIX. Look at the screenshot below from OptionNet Explorer (ONE):

As seen in the Part 1 table, this number is 97.33 and 102.89 on 3/12/20 and 3/16/20, respectively. This website says the highest closing VIX ever recorded was 82.69 on 3/16/20, which therefore means the highlighted number can’t be VIX. Granted, I use data from 3:55 PM rather than the close for this backtest, but I find it hard to believe VIX could come down 20 points in five minutes. Until told otherwise, I am therefore calling the highlighted number “SPX IV” (probably calculated via proprietary ONE algorithm) rather than “VIX.” I will contact ONE Support to get a more complete explanation (also on what the percent change is based because 4:00 PM the previous day is not correct).

The Part 1 table says the WH performed miserably through the March 2020 market crash. Backtrading 10 contracts weekly of this strategy lands us down $650K on 3/23/20!

Per usual, I tried to backtest realistically by using values only 20% of the range from natural to midprice. While most backtesting chooses to ignore it, slippage is a reality of trading. We do not want to be the camp described in this second paragraph, else we’re probably better off finding a different strategy altogether. As you can imagine, when the market speeds up the spreads widen and I sometimes had to go nearer to the money to make sure I got requisite PCS net credit. As a single, inexpensive ($0.45) leg, buying the extra long is not affected by this so much.

We know when the market crashes, the PCS are going to “get killed.” This is explained in the WH video series and can certainly be seen in the table. The PCS studied started 20.9% and 22.5% OTM, which was not enough distance to keep them safe.

The video says when the PCS gets killed, the long puts will explode to cover them; this is where I run into problems. I sell a PCS every week (risk) and I buy two additional long puts (insurance) as well. This makes the position seem safe. One put is a 7 DTE teeny that serves only to keep T+0 elevated (saving portfolio margin). Even without the teeny, the second put leaves us net long contracts. What could possibly go wrong?

You sleep on that, and I will continue the discussion next time.

Debunking the Williams Hedge (Part 1)

One does not have to look far to find trading courses (or systems) for sale. The advertising pitches are as long as the Mississippi River (or a CVS receipt, at least). I have written about this here and in the second-to-last paragraph here. Today I am going to give a real example of why you might want to avoid purchase of vendor strategies without testing them yourself.*

Submitted for your approval: the case of one Williams Hedge (WH). With name changed to protect the innocent, this is an option hedging strategy sold for ~$300 by a company formed in 2016. The description reads:

     > The WH is designed to be a costless way to add crash protection…
     > this trade will not help in mild corrective markets… this trade
     > will kick in hard in a crash and offer profit to offset losses as
     > well as significant margin boost to counteract exploding margin
     > requirements and forced liquidations at inopportune times.
     >
     > [Our company is] an alternative to the hundreds of websites offering…
     > index [ICs advertising] 5-10% every month without mention of the
     > occasional 80% loss… [the WH is] different… adaptable, robust,
     > tested… [we] backtest the idea and rules, then vet the trade plan
     > through a small team to tear it apart, then backtest it some more,
     > and then create a course for it while we trade it live, and then we
     > launch. If it doesn’t perform to spec, we scrap it.

Sounds pretty solid, right?

The WH strategy includes:

• Long-dated put credit spread (PCS) sold for around $0.60
• Short-dated long put (LP) at short PCS strike for $0.45 or less
• 7 DTE teenie for $0.05 as margin control
• Rinse and repeat every week

The video says the far-dated PCS is going to get killed in a market crash while the near-dated LP will explode to offset losses and then some. Gains on the PCS won’t be realized until it closes. During market declines, the overall position will drawdown due to unrealized PCS losses but will recover by expiration assuming the PCS remains OTM.

Parental discretion is advised for the following:

This is an ~32% loss in the S&P 500 (SPX) index in one month. We have isolated a true Black Swan in its natural habitat.

What you are about to see is nothing short of phenomenal:

The second row of the table corresponds to the red arrow and highlighting on the chart. The Cum(ulative) % Chg columns use this row as base values for SPX IV and SPX. The three columns of PCS data are spread price, days to expiration, and moneyness. Red values indicate ITM spreads.

I really needed to immerse myself in these numbers for a while to get a good sense of what I’m looking at. SPX falls 30.9% while IV increases ~608%. SPX down 59 points (-1.79%) on 1/31/20 is a 2.48 SD selloff while the 106-point selloff (-4.38%) on 3/20/20 is only 1.29 SD. Realize we’re dealing with IV ~11 in the former case and IV ~74 in the latter.

I will continue next time.

* — See what I did there? Even after payment, many vendors will not be transparent with the
       strategy (or coding) much less let you test all the rules yourself. You certainly won’t be
       able to fully backtest a strategy before paying for it.

Call Me Crazy (Part 9)

Today I finish up tying together loose ends with regard to the long call (LC) backtest.

In the bulleted list here, I suggest we might vary the DTE/DIT ratio. The current backtest holds two-year LEAPS for one year. What if I hold for 18 months? What if I buy 18 months out and hold for one year (plus one day)? Holding longer keeps the number of transactions as well as taxes down (LTCG but please consult a tax advisor as I am not a CPA and do not know your individual situation). One tradeoff is time decay, which increases as the LC gets closer to expiration.

One way to get more data points in the backtest is to buy a shorter-dated LC and hold closer to expiration. The original backtest has only 15 data points using two-year LEAPS. Alternatively, I could buy the LC one (two) month(s) to expiration and roll after three (six) weeks, for example. Many permutations are possible and what I would ideally like to see for a robust strategy is most to be profitable with superior risk-adjusted returns (especially in lieu of accelerated time decay).

Holding the LC longer when the market has declined can improve risk-adjusted return. LCs would be held for varying durations depending on underlying share performance. For a multi-year down market, this idea really seems attractive because a max loss could only be realized every two years rather than every year. Thinking about 2008-9 in the original backtest, maximum drawdown (MDD) would have been closer to 20% than the 32% seen in the enhanced data set.

When doing this “mental stress testing,” always think about the pros and cons of different tweaks. Holding the LC for longer lowers MDD in an extended downturn at the cost of slower profit after upside reversal. The unrolled LC expires worthless if the underlying fails to reclaim the strike price by expiration whereas the rolled LC profits earlier despite a larger MDD.

Also when doing “mental stress testing,” be aware that small sample sizes are very susceptible to curve fitting. We can imagine specific historical periods and forecast hypothetical performance, but when broken down into metrics (e.g. severity, duration, volatility change), each DD period is unique and small sample sizes are not to be generalized (see second-to-last paragraph here for fallacy of the well-chosen example). In the end, the best answer may be to skip the backtesting altogether and do what feels right because aborting a trading plan midstream is a good way to lose money and make us not want to look back.

Is rolling the LC up a low-probability trade and therefore something to avoid? This is effectively a bear call credit spread. Stocks have an upward bias while a call credit spread is bearish/neutral. Failure to roll would result in a more expensive LC over time. Perhaps fewer contracts could eventually be implemented to [somewhat] smooth out asset allocation.

Sticking with the same theme, rolling the LC down—effectively a bull call spread—should be a high-probability trade. This suggests we roll down when the market tanks and hold the strike as it soars. I tend to agree that when rallying, we want the LC to mimic stock. If we don’t roll, though, then we can’t get downside protection. I’m not sure how to rectify that.

The challenge with investing may not be figuring out the right or wrong strategy as much as determining the pros and cons of each and making educated choices from there.

Despite the benefits, one disadvantage to the LC is losing money if the underlying is flattish to down. This sounds like good reason to try lowering its cost basis, which is where I will head going forward.