Dynamic Iron Butterflies (Part 5)

Better execution can definitely improve profitability of the dynamic iron butterfly (DIBF). This makes me wonder about placing the trade and just leaving it to sit until filled.

Analysis of the maximum adverse excursion (MAE) can help to study this. A losing trade could almost always have been filled at a better price. Out of 3968 backtrades, 736 never get down more money than they are at trade inception. Minute-to-minute volatility in the markets is much greater than day-to-day (seen in my EOD backtest) volatility, which provides hope that most of the 736 would eventually fill. This is no guarantee, however, and 18% includes a big opportunity for otherwise winning trades to go unfilled thereby depressing total return.

Here is the complete MAE distribution across 3,968 backtrades:

The x-axis indicates the maximum MAE seen for each particular grouping of trades. The first data point includes all trades with MAE less than 10%, the second data point includes all trades with MAE of 10% up to 20%, etc.

The graph illustrates a majority of trades never get down a whole lot.

Here’s the same information packaged somewhat differently:

With a profit target of 10%, over 64% of all backtrades never get down more than 20% and over 86% never get down more than 50%.

This suggests a potential benefit to using a stop-loss. I still need to better understand how many winners would be affected, though. How does the MAE distribution compare between winning and losing trades?

This graph shows the difference to be significant. Over 91% of winners never get down more than 30% compared to just under 13% of losers. The average MAE for winning (losing) trades is 12.9% (57.9%).

Over 98% of winning trades are never down more than 50%.

Is a 50% stop-loss worthwhile?

The cost is conversion of 59 winning trades into certain losers ranging anywhere from 50-90%.

The benefit is an opportunity to save money on 467 trades that would otherwise lose up to ~100%.

This sounds like a reasonable trade-off to me since the number of trades potentially benefitting is eight times the number to suffer a worse fate. And even if the overall return does not improve, cutting down the biggest losers would still reduce the standard deviation of returns, which is a measure of risk.

Confirming this would involve collecting revised loss data on 526 backtrades with MAE over 50%. I should also backtest the surrounding parameter space by checking stop-loss thresholds like 10%, 20%, etc. This could be an awful lot of work…

But who ever said making a business out of trading should ever be easy?

Dynamic Iron Butterflies (Part 4)

My last post was a hodgepodge of information about the dynamic iron butterfly (DIBF) backtest and I’m going to pick up right where I left off.

The differential loss data compounds my concern about the backtest being a combination of two different trades. As a combination, it’s hard to tell to what extent the “dynamic” [put credit spread (PCS)] aspect contributes to the profitability. Worrisome is the fact that 77% of the losses are to the downside: exactly where the “dynamic” aspect can result in greater loss. Let’s take a closer look at the loss distribution:

Each data point represents what percentage of upside (green) or downside (red) losses occurred at different magnitudes of loss (ROI %). The difference does not look as dramatic as I expected but it still may be. A greater percentage of downside losses occurred at loss levels exceeding 70% ROI. Furthermore, 66% (51%) of downside (upside) losses were worse than 40%. This is a rather large difference especially considering the fact that 77% of all losses were to the downside.

Plotting number of losses, rather than percentage, looks like this:

The difference looks more dramatic here. Total dollars lost would mimic this.

While the PCS is an obvious culprit when studying directional loss distribution, we cannot forget that it benefits the position to the upside. More than just “upside,” the PCS benefits the position as long as the market does not fall more from the DIBF short strike than the width of the call credit spread. This is the “dynamic” portion. I don’t need to do the computations to know this is the vast majority of winners. In fact, this will always be the case:

If the market falls an amount equal to the width of the call credit spread, this trade will be losing money as evidenced by the intersection of the red horizontal hashmark and blue line being below zero profit (circled in red on the y-axis). That would be my reason for hypothesizing all winners will have fallen less than this amount even though before expiration, unrealized loss is not as extreme (e.g. purple curve).

In any case, what would be interesting to know is how DIBF performance compares to a symmetrical iron butterfly (no PCS). This is definitely a future direction for study.

Next on the docket, though, is analysis of maximum adverse excursion.

Dynamic Iron Butterflies (Part 3)

In the last post I suggested overstatement of transaction fees may have been the difference between a winning and losing dynamic iron butterfly (DIBF) backtest.

I crunched the numbers and can now speak definitively to that hypothesis. Here are the overall backtested trade statistics recomputed for different transaction fees (TF). The second column is the $0.26/option used in the original backtest:

Indeed, lowering the transaction fees does make this a profitable trade. While the profit factor of 1.15 is not nearly as impressive as the 1.58 seen for naked puts, it is at least profitable.

Put another way, these numbers provide reasonable doubt as to whether the DIBF is actually a losing trade. My preference is still to bias slippage in favor of loss and regardless of the heavy slippage applied, I want to see if I can do anything to boost the average trade and make this strategy more encouraging.

Doing the backtest helped me discover two potential problems.

First, the DIBF has a varied reward-to-risk ratio (RRR). I noticed at times the RRR was downright rotten (less than 1.0). This could be improved by decreasing the width of the put wing. I suspect a lower RRR might lead to a lower winning percentage and a longer time in the trade because the T+0 curve is particularly steep to the downside in relation to the limited potential profit at expiration. RRR was not tracked in the current backtest so my suspicion remains speculative.

My second issue with the strategy is the fact that it is a combination (of an iron butterfly and put credit spread). Selling an OTM long put and buying one further OTM—a credit spread—is what makes it “dynamic.” Before delving too much into the results of a combination strategy, I feel inclined to first study a plain iron butterfly. I may or may not do that before having the confidence to trade the DITM regularly.

Thanks to Pete_UK for helping me calculate the directional breakdown of losses in a hurry:

This compounds my concern about adding a put credit spread to the iron butterfly.

Dynamic Iron Butterflies (Part 2)

Last time, I presented the overall trade statistics for my first study of dynamic iron butterflies (DIBF). The results were not pretty. Today I want to address the impact of transaction fees (slippage + commission).

I have previously discussed how transaction fees can make or break a study. I subtracted $0.26/contract because I was backtesting some expensive, at-the-money options and when I have to estimate, I prefer to bias in favor of loss. Despite only trading a handful of live butterflies to date, I have never paid more than $0.13/contract in transaction fees (sometimes $0.06). Fast-moving markets could take more than $0.26 but such adverse conditions are rare.

Let’s compute how an overestimation of transaction fees may have affected results. The mean margin requirement (MR) across all trades was $4,878. I subtracted $26 * 8 = $208 from each trade for transaction fees. Cutting that by 50% adds $104 / $4,878 * 100% = 2.1% ROI to each trade. The average trade lost 1.4% so this modification makes the backtested DIBF a winning trade (+0.7%). If even 50% is estimating high then I should reduce transaction fees between $104 – $156. If I use the middle of that range then the average trade gains 1.2%.

Transaction fees alone can make the difference between a 1.4% loss and a 1.2% profit per trade. Financially speaking, those are worlds apart. Thinking about how many traders omit transaction fees entirely for the sake of simplicity just boggles my mind. No wonder so many statistics suggest up to 90% of traders fail within the first five years.

These calculations are based on averages but the exact MR should be considered. Reducing slippage by $52 impacts a $2600 trade twice as much as a $5200 trade. More expensive trades occur in higher IV, which occurs less frequently. I would therefore hypothesize lower MR trades to dominate the distribution, which might boost average ROI further:

Indeed, 88.8% of trades had MR within the lower half of the range. While cheaper trades significantly outnumbered expensive ones, only 54.8% of trades had MR below the arithmetic mean. The impact of this skewed distribution is questionable.

Having a more significant impact would be trades where the long option(s) would have been left to expire worthless in live trading. These seemed to occur later in the data series at higher prices for the underlying. Being forced to allow one or two long options to expire worthless saves approximately $20 or $40 per trade, respectively.

I will continue the discussion in my next post.

Dynamic Iron Butterflies (Part 1)

I want to trade butterflies and the only way I can get myself to trade something new is to backtest it. This study is based on a Tasty Trade segment from April 1, 2016.

The subject of this analysis is dynamic iron butterflies (DIBF). Without further ado, let’s jump ahead to the results:

This is a completely different profile than that presented in the Tasty Trade video! Make no mistake: with an average loss posted after 3,900 trades, this is ugly.

These negative results were very surprising to me. I have been conditioned to be a believer in short strangles (the butterfly’s undefined-risk counterpart) and I have personally done some backtesting to support that belief.

Before throwing the proverbial baby out with the bathwater, let’s step back and critique the methodology.

My first thought is that I was probably heavy on the transaction fees. $26/contract might be reasonable during fast-moving markets but is probably excessive in most cases.

Second, instead of expiring DOTM longs in the later years of backtesting, I sold them for the nickel (or less) they were stated to be worth. In reality, I probably would not have been able to sell them so close to expiration and I would have been spared that $26/contract. Proceeding in this fashion saved time (it took me four months to do this study) and I am typically comfortable with backtesting bias that favors the losing side.

My third question mark surrounds asymmetrical loss, which may or may not be an issue. I calculated profit/loss in terms of ROI(%) because the margin for a 1-contract trade ranged from $1,401 to $12,400. With ROI(%) itself serving as normalization, I discovered the discrepancy. In some cases the market moved far against the trade to the downside causing > 100% loss (transaction fees). In other cases the market moved far to the upside causing a more limited loss (e.g. 50 – 90%). Some trades also ended up symmetrical: maybe these should be separated out?

The average loss was just over four times the average win, which completely nullifies the benefit of 78% winners. That suggests an MAE analysis to see if stops could be beneficial.