Test 7: IV/HV

I thought it would be fun today to test whether it makes sense to use implied volatility as an entry criteria to our tests. Specifically, I am looking at whether the ratio of IV to HV (historic vol) will be beneficial.

I’ve heard that generally speaking it is a bad practice to sell premium when the HV is larger than the IV. We like to sell in high IV environments with less underlying movement (actually that is ideal). When IV is lower than HV, can we get enough premium to justify the potential large swings in the underlying? We are actually under that situation right now with the VIX at 13.49.

For this test, I construct rules that only allow trades to be entered when the IV/HV ratio is greater than (1.0, 1.1, 1.25, 1.5, 1.75, and 2.0). No other entry or exit parameters are used. Note that in this test I am only using the implied volatility of the front-month option chain calculated based on a weighted average of the ATM call and put IVs.

Results

1 Month 2 Months 3 Months All
Improved PF 9(17%) 38(70%) 38(70%) 85(52%)
Improved MaxDD 54(100%) 54(100%) 54(100%) 162(100%)
Improved ExpRet 9(17%) 38(70%) 38(70%) 85(52%)

The initial pass is interesting; the back months show a much higher benefit that the front month. The heatmap shows this in even more detail:

iv-heatmap-diffpf

The obvious conclusion is that if you find yourself with a 2x IV over HV, make the trade, but the data shows these are extremely rare (i.e. 6 trades in 9 years). Discounting these trades, we can see that generally speaking in the other cases using IV/HV as a marginal impact.

There are several trades that really stand out to me, all in the 2nd month. The 10-Delta, 15-Delta, and 18-Delta trades with IV/HV of 1.25 and 1.5 all saw at least 25 trades during this period and had substantial improvements. The 10-Delta 1.5 IV saw only winners. What is so special about this cluster of trades? Digging into the specific trades they show that they only have a few losses, but their neighbors have about the same number of trades and just 1 or 2 more losses. Those extra losses were enough to dramatically skew the results considering this is an expiration test and the losses were all max losses.

Now, what happens if we also include the early exit trigger from the previous test?

iv-style-heatmap

In this case we start to see pretty consistent and significant improvements in the back month trades. These are also fairly significant improvements over the early exit tests themselves. And note the cluster in the second month trades is no longer as significant.

Conclusion

On its own, the IV/HV ratio is not a great indicator to help enter trades. While there are scenarios where the ratio seems to help a great deal, these come at the cost of significantly less trades. However, it seems that paired with other triggers it may be more beneficial than not.

The higher the IV/HV ratio, the less number of trades that are going to occur. Is it worth it to wait until the IV is greater than the HV, in my opinion… yes if you are looking at back months. I am actually quite surprised that there was not a more dramatic effect for the front month since this test was using the front-month IV in its calculation. In the last 9 years there are only a few cases where the IV/HV ratio was less than one (i.e. HV greater than IV), and only one period where the HV was higher for sustained period of time.

Overall, I’m not convinced this tool is critical unless one is already in a low vol environment. If the HV is high, but the IV is also high, we would expect to be paid a decent premium and be able to get wider strikes, possibly nullifying the effect.

It is also interesting to note that the HV is a lagging indicator and the IV is a current sentiment indicator based on the market’s perception. Realized volatility in the past may not carry though to the time when HV is greater than IV. This would result in potentially missing trades good trades, and possibly allowing some bad ones.

Test Stats

Permutations 162
Profitable Scenarios 121 75%
Total Trades 7283
Profitable Trades 5212 72%
Expired 2442 34%
MaxDays 4841 66%
Test Duration 1.38(min)

Test Files

Excel Data

Advertisements

Test 6: Early Exits

Today I am going to test early exits. To date, my tests have held trades through to expiration unless a target or stop loss was reached. But many traders prefer to exit trades well before the expiration to minimize gamma risk. Of course, one would miss out on the largest theta decay, but theta and gamma often are opposing greeks.

The general sense I get is that those entering front month trades attempt to hold though to expiration, possibly exiting with 1 week left. Those entering the back months tend to be out of the trade with at least 3 weeks left to go. This run will test exiting 1 through 4 weeks prior to expiration, with no other entry or exit parameters. With this test, I am simply hoping to find a best practice depending on the expiration month.

Results

1 Month 2 Months 3 Months All
Improved PF 18(50%) 35(97%) 28(78%) 81(75%)
Improved MaxDD 36(100%) 36(100%) 36(100%) 108(100%)
Improved ExpRet 12(33%) 24(67%) 26(72%) 62(57%)

The table shows that where were substantial improvements in the Profit Factor for the back month. In all cases the risk was reduced, but in the majority of cases in the back months the expected returns were also improved.

Looking at a heatmap for the improvement in profit factor over the baselines:

EE-heatmap-diffPf

We can wee that front month actually saw an improvement when the trades were exited 7 days prior to expiration. And very high delta trades may also see an improvement at all exit periods. However most other trades saw marginal or negative improvements as we might expect from these trades.

The back month trades are another matter. Clearly exiting early has a benefit on nearly all these trades. Exiting 3 weeks prior seems to be a general sweet spot for both back month trades, however, the 3-month 10-Delta trade saw the largest improvement when held to the last week.

So there is something to exiting early. Lets see if we can nail down some better rules. I reran the test, but this time included all days between 3 and 28 prior to expiration.

HeapMap-EE-3-28

With this heatmap I removed all the weekend days and then color coded each row individually. Green reflects the most improvement and red the worst by delta. Red does not necessarily mean bad as any value over zero is an improvement over the baseline. The aim is to show the best exit point for each expiration month and delta. For example, if you were to enter a trade with the front month and a 30 delta, an exit with 21 days remaining may actually be in your interest.

What is interesting about this map is that several patterns emerge.

  • Exiting 22 days for the 3 month trade is generally optimal except for the delta-10 trades where holding to 4 days before seems to be beneficial. This is very anomalous.
  • Exiting between 17 and 22 days for the 2 month trades has the highest benefit.
  • Exiting early in the front month generally has marginal to negative improvement. Low delta trades may benefit from exiting just prior to expiration. Otherwise they are harmed. Higher delta trades show a marginal benefit to exiting early at most points with the largest gain being 17-22 days prior (note these would be very short term trades).
  • Exiting 2 weeks prior on the Friday seems to perform much worse than if you were to exit a few days before or after that point. Why two weeks before? This pattern is not seen at 3 or 4 weeks on fridays.

Now, why is there an abnormality around the 3-Month, 10-Delta, Day-4 exit trade? It performs significantly better compared to its neighboring trades. The neighboring trades on the same delta have identical entry points and the same number of trades. However, when exited one day earlier (day 5) the scenario wound up with 4 more losses. Similarly, the neighbor in the 15-delta row had a few more total trades, but also experienced 4 more losses. Given that many of the trades were identical and that just exiting one day vs another made such a difference in 5% of the trades, I suspect this is just a statistical anomaly that would be smoothed out over time with more trades/data.

Conclusion

Clearly there is a benefit to exit back-month trades earlier. It substantially decreases the risk of the trade, and in many cases leads to higher expected returns probably by missing out on larger losses. Exiting between 18 and 22 days appears to lead to the largest benefit.

For most of my tests going forward unless noted otherwise, I exit trades for back month trades based on this data.

Test Stats

Permutations 702
Profitable Scenarios 584 83%
Total Trades 141185
Profitable Trades 91969 65%
MaxDays 141185 100%
Test Duration 1.56(min)

Files

Excel Data

Test 5: Targets, Stop Losses and Min Premium

Now its going to get more interesting. We combine Profit targets, Stop Losses and Minimum Premiums together to see if combined they are greater than individually.

For this test, I am going to use the same Profit Target and Stop Loss parameters in the past tests. I am also going to use the historical average premiums as an entry point into the trades (i.e. the minimum premium is set to the average historical premium).

High Level Breakdown

The breakdown suggests that once again the front-month trades are harmed by these trade parameters, but the 2nd and 3rd month trades find improvements.

1 Month 2 Months 3 Months All
Improved PF 211(23%) 558(62%) 657(73%) 1426(53%)
Improved MaxDD 900(100%) 900(100%) 900(100%) 2700(100%)
Improved ExpRet 218(24%) 359(40%) 553(61%) 1130(42%)

Profit Factor Improvement

A heat map of the profit factor improvement (over the baseline) shows much more detail. Generally speaking, all parameters decrease performance for front month trades. The high delta trade for the front month does show marginal improvement consistently, but not enough for me to be confident.

TSP-Heatmap-PF

The second and third month trades are a different story. It is hard to find an exact pattern, but we can claim that generally adding these parameters improves performance of these trades, with substantial gains for lower delta trades (i.e. Delta 10). The Delta 10 trades in the second month and some of those in the third month have massive improvements in performance compared to the baseline. Ideally I would like to see the neighboring trades (i.e. Delta 15) also experience similar but diminished gains in order to point to a possible pattern, but that is not the case (it could be that my resolution is not fine enough and that Delta 12 trades show that diminished improvement).

Looking deeper at the specific trades of one neighboring trades (all things being the same except the delta), its trades experienced 6 losses instead of one. This may add to the credibility that the low delta trades improvement was not just due to luck.

Expected Returns

A heatmap of the expect returns of each scenario shows that generally speaking

  1. High Delta trades still perform poorly regardless of month
  2. Very Low target trades perform poorly,
  3. Low stop loss trades perform worse
  4. Front Month trades perform best when the target is > 70%, Deltas <= 25, and stop losses > 30%
  5. Second and Third Month trades perform best when Target >= 30%, Stop Losses >= 20%, and Deltas <= 25%

TSP-Heatmap-ExpRet

We can also look at the top three trades for each expiration period so gain similar insights.

Month Delta Target StopLoss ExpRet MaxDD PF WinRatio Trades AvgDays
1 1st 0.10 90% -90% 6% -127% 2.2 93% 89 20.8
2 1st 0.10 90% -100% 6% -127% 2.2 93% 89 20.8
3 1st 0.10 90% -60% 6% -127% 2.2 92% 90 20.4
1 2nd 0.10 40% -60% 6% -71% 10.4 99% 115 19.1
2 2nd 0.10 40% -70% 6% -71% 10.4 99% 115 19.1
3 2nd 0.10 20% -70% 3% -71% 8.7 99% 180 10.1
1 3rd 0.10 20% -70% 3% -78% 7.1 99% 151 15.2
2 3rd 0.10 20% -80% 3% -100% 5.5 99% 151 15.2
3 3rd 0.10 20% -90% 3% -100% 5.5 99% 151 15.2

Those are some impressive win ratios. Those with a 99% win ratio only experience a single loss, and it was during the Oct 2008 market crash.

It is also interesting to note that as the expiration period increases, the expected return decreases by so does the average duration of the trade. Trades with shorter duration may be sitting on the sidelines frequently.

Conclusion

The combination of targets, stop losses and minimum premiums seems to substantially improve low delta trades in the second and third month, and to a lesser degree most of the other trades in those expiration period. But again, while the risk in the front month trades was reduced, the overall profit factor of almost all trades was also reduced.

Test Stats

Permutations 2700
Profitable Scenarios 1745 65%
Total Trades 526411
Profitable Trades 229165 44%
Expired 59983 11%
StopLoss 48866 9%
TargetMet 417562 79%
Test Duration 2.60(min)

Test Files

Excel Test Results

Test 4: Minimum Premiums

This test will use minimum premiums on trades to determine whether this simple addition to our entry criteria can generate alpha.

Adding minimum premiums is actually a little complicated as the question is really relative to what. The first part of this test will blindly accept trades if the trade is over a given premium. The variety of premiums I use are:

1.0, 1.25, 1.5, 1.75, 2.0, 2.25, 2.5, 2.75, 3.0, 3.25, 3.5, 3.75, 4, 4.25, 4.5, 4.75, 5.00

This creates 459 different scenarios for the three different expiration period. Now some of these premiums don’t make sense and/or result in duplicate scenarios. For example a $4 premium on a front-month trade with a delta of 10 is pretty hard to come by (we found none). But in the analysis, I will remove scenarios that don’t meet a minimum of 25 total trades.

Summary Breakdown

1 Month 2 Months 3 Months All
Improved PF 17(17%) 14(13%) 18(16%) 49(15%)
Improved MaxDD 100(100%) 107(100%) 112(100%) 319(100%)
Improved ExpRet 22(22%) 15(14%) 19(17%) 56(18%)

The summary shows that again we find the adjustment reduces the risk in all cases. Surprisingly the addition of a minimum premium does not generally help the expected return, but again, we are looking at many unrealistic trades. So we need to dig deeper.

Many of the trades actually matched the baseline. That is the minimum premium was low enough that all trades found in the baseline test, matched the scenario. If I strip out those trades and we look at a heatmap of the improvement in the Profit Factor we can see that the PF for most trades is marginally affected.

mp-heatmap-diffPf

Looking deeper still, the average number of trades for this test is about 20% smaller than the baseline. Since the baseline tests actually performed well on average, this minimum premium threshold is preventing us from participating in more successful trades (and potentially allowing negative ones when the volatility of the market is high).

Average Premiums

So using random minimum premiums was a wash. I wonder whether using minimum premiums based on the average premium for each delta and period would be of benefit. I wrote a program to collect the average premiums for all iron condors found for all deltas between 4 and 40 on every trading day in the first 3 months. For example, the program would find the average price for a RUT IC entered with a delta of 12 with 56 trading days left till expiration.

ap-avgprem

With the exception of the very last day of trading where premiums appear to perk up, these patterns resemble the theta decay we would expect. I double checked the premium on a delta 40 trade today (the last day of trading in Dec expiration) and it matches the chart above. The tests don’t initiate trades anywhere near the last day, so this anomaly is ignored.

Running the test against the average premium results in only 27 permutations  but doesn’t yield any great advantage either:

1 Month 2 Months 3 Months All
Improved PF 2(22%) 1(11%) 3(33%) 6(22%)
Improved MaxDD 9(100%) 9(100%) 9(100%) 27(100%)
Improved ExpRet 3(33%) 1(11%) 3(33%) 7(26%)

mpa-heatmap-diffPF

Again, the change in PF is overall negative, but the changes are generally marginal.

I didn’t really expect using the average price to help alot since it is after all – average.

Average + 1 SD

Lets try for something much better than average. I take the average price plus 1 standard deviation. This means that the minimum price should be better than 84% of the historical prices.

ap-sdprem

The standard deviation as a fair amount of noise to the prices, so I am actually going to smooth it out using a very small moving average. This still results in 27 scenarios.

1 Month 2 Months 3 Months All
Improved PF 6(67%) 2(22%) 3(33%) 11(41%)
Improved MaxDD 9(100%) 9(100%) 9(100%) 27(100%)
Improved ExpRet 7(78%) 2(22%) 4(44%) 13(48%)

Now we are starting to finally see some improvements, but they are mostly in the front month trades. But again, the changes are all marginal.

mpasd-heatmap-diffpf

Conclusion

I would have thought that adding minimum premiums to our trade would improve the performance but it seems to keep us in most of the loosing trades and away from many of the other winning trades. The average number of trades is reduced in most scenarios, and is substantially reduced in some to the point that we consider them too small to rely on.

I tried several variations of setting minimum premiums and all seemed to lead to marginally better or worse performance. Every trade helped reduce the risk, but the offset in lost expected return reduced the performance further.

Summary Stats

Permutations 459
Profitable Scenarios 190 41%
Total Trades 30368
Profitable Trades 19709 65%
Expired 30368 100%
Test Duration 1.38(min)

Test Files

Minimum Premium Test Data

Average Premium Data

Min. Premium using Historical Avg Test Data

Min. Premium using 1 SD Test Data

Test 3: Targets and Stop Losses

Profit targets were found to be helpful, and Stop Losses less so, but can the combination of the two be greater than the sum of the parts?

Test Setup

For this test I create scenarios that include all profit targets and all stop losses from the previous two tests. Targets and Stop losses from 10% to 100% will be used. Ultimately this creates 2700 unique scenarios across the first three months.

Now that we have four input parameters (Delta, Days to Expiration, Profit Targets, and Stop Losses) for every scenario, analyzing the general trends is much harder. Ideally I need a holographic projection to show the data in three dimensions – but the budget here at Option I/O is low, so we will have to make due.

Highlevel BreakDown

We start by looking at the high-level break down of the tests.

1 Month 2 Months 3 Months All
Improved PF 200(22%) 645(72%) 713(79%) 1558(58%)
Improved MaxDD 900(100%) 900(100%) 900(100%) 2700(100%)
Improved ExpRet 202(22%) 441(49%) 573(64%) 1216(45%)

This are pretty surprising results to me. There was no question that adding stop losses and targets limits should improve the risk for every trade since we saw that in the last two tests. But I thought the combination of the two would help with the over all expected return as well. But what we see is the Front Month trades are resoundingly impacted by adding stops and limits, however the 2nd and 3rd month trades have a definite improvement.

Looking at a heatmap of this data shows more detail.

This heatmap is a little complicated to read. The major columns are the expiration months of the trades, the minor columns are the deltas. The Major rows are the profit targets, and the minor rows the stop losses.

This heatmap is a little complicated to read. The major columns are the expiration months of the trades, the minor columns are the deltas. The Major rows are the profit targets, and the minor rows the stop losses.

This data does not mean that any of these trades are necessarily bad trades. This is simply a comparison to the baseline test. The red indicates the trade is worse than holding to expiration based on the Profit Factor. Green is an improvement.

Here are some of the takeaways from this data:

  • Front-month trades are heavily impacted until the profit target is ~100%. Even then adding stop losses seem to help some, but not all the trades. And most gains or losses are marginal.
  • Almost all CTM trades in the 2-3 month time frame were improved.
  • Very low stop losses are stopped out a lot and are generally detrimental to the trade.

Looking at the heatmap in another way (not posted here but in the data file), I see a rough pattern where in the 2-3 month trades, the largest benefit is received by high delta/low target trades though to low delta/high target trades. There seems to be a sweet spot depending on the delta of the trade. You can see in the data that as the delta is decreased the sweet spot moves.

One final point in this section, the largest benefit goes to the CTM trades. This can easily be seen in the following chart. The DiffPF is the improvement in the Profit Factor over the baseline.

TS-diffPF-vs-delta

Expected Returns

With 2700 points, this is also a little hard to read, but this graph shows the general trends based on the delta and month of the trade. Most Delta 10 through 20 trades appear to have a positive expected average. While most with 30 and above do not.

TS-expRet-delta

Top 3 Trades for each Month

The following table shows the top three trades (by Profit Factor) for each of the different expiration months.

Month Delta Target StopLoss ExpRet MaxDD PF WinRatio Trades AvgDays
1 1st 0.10 90% -90% 6% -100% 2.5 95% 112 21.7
2 1st 0.10 90% -100% 6% -100% 2.5 95% 112 21.7
3 1st 0.10 90% -70% 5% -100% 2.2 94% 112 21.6
1 2nd 0.10 90% -40% 7% -100% 2.7 92% 102 41.7
2 2nd 0.10 90% -80% 7% -100% 2.7 95% 100 43.2
3 2nd 0.10 90% -90% 7% -100% 2.6 95% 100 43.2
1 3rd 0.10 30% -90% 3% -100% 3.3 99% 136 24.9
2 3rd 0.10 30% -100% 3% -100% 3.3 99% 136 24.9
3 3rd 0.10 60% -90% 6% -100% 3.2 97% 90 44.6

What is interesting to note is that many of these trades are very close to the baseline test. It’s also interesting to note that all the trade are with a delta of 10. Personally I would have thought that all the top trades for any month would have been slight variations of each other, but two trades stand out as different. The 2nd month 90/40 trade and the 3rd month 30/90 trade.

Top 3 Trades by Trading Style

I thought it would be interesting to also start listing the best trades by style.
(See my previous article about trader styles here.)

Month Delta Target StopLoss ExpRet MaxDD PF WinRatio Trades AvgDays
FrontMo-FOTM 1 1 0.10 90% -90% 6% -100% 2.5 95% 112 21.7
FrontMo-FOTM 2 1 0.10 90% -100% 6% -100% 2.5 95% 112 21.7
FrontMo-FOTM 3 1 0.10 90% -70% 5% -100% 2.2 94% 112 21.6
FrontMo-CTM 1 1 0.15 100% -30% 8% -112% 2.0 82% 113 23.3
FrontMo-CTM 2 1 0.18 100% -40% 9% -120% 1.9 79% 114 23.1
FrontMo-CTM 3 1 0.15 90% -30% 6% -124% 1.8 82% 114 21.9
BackMo-FOTM 1 3 0.10 30% -90% 3% -100% 3.3 99% 136 24.9
BackMo-FOTM 2 3 0.10 30% -100% 3% -100% 3.3 99% 136 24.9
BackMo-FOTM 3 3 0.10 60% -90% 6% -100% 3.2 97% 90 44.6
BackMo-CTM 1 2 0.15 90% -70% 11% -200% 2.4 92% 103 47.4
BackMo-CTM 2 2 0.20 30% -40% 6% -100% 2.4 91% 141 22.1
BackMo-CTM 3 2 0.18 50% -40% 8% -141% 2.4 88% 116 29.2

Again we see that for Front-month trades, the best scenarios are the same or close to the baseline. The back-month CTM trade however shows there may be an advantage with a 50% target and a 40% stop.

Conclusion

Surprising, Profit Targets and Stop losses do not have a major impact on the front month trades, but they do on the 2nd and 3rd month trades. They reduce the risk in every scenario. There is a sweetspot for the stop loss, but it is dependent on the delta of the trade.

Test Stats

Permutations 2700
Profitable Scenarios 1845 68%
Total Trades 679984
Profitable Trades 298187 44%
Expired 65216 10%
StopLoss 66814 10%
TargetMet 547954 81%
Test Duration 2.57(min)

Test Files

Excel Data

Test 2: Stop Losses

What level of stop losses are effective for managing iron condors? Our baseline test effectively had a 100% stop loss on all scenarios. Will adding stop losses based on the at risk margin (See the test on Profit Limits to learn why I use this metric) improve the performance of Iron Condor trades?

My guess before looking at the data is that stop losses will be shown to drastically reduce the maximum risk, and should be able to improve the effective expected return. Because some tests will be stopped out quiet frequently (i.e. those with a 10% stop loss threshold), I expect many tests to have a higher number of trades compared to the baseline.

Test Results

The test scenario use stop losses on the margin at risk for the following values:

10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 100%

The following summarizes the performance of the test run:

1 Month 2 Months 3 Months All
Improved PF 32(36%) 28(31%) 28(31%) 88(33%)
Improved MaxDD 90(100%) 90(100%) 90(100%) 270(100%)
Improved ExpRet 34(38%) 26(29%) 25(28%) 85(31%)

As expected the risk was substantially reduced in all cases. The vast majority of scenarios actually experienced lower expected returns and lower and lower profit factors. Meaning that the gains in risk were still outweighed by the loss in expected returns.

Profit Factor Improvement VS Delta

Looking at a histogram of the difference in profit factor compared to the baseline, we can see some patterns, but its not as clear cut. Very small stop losses perform poorly on average. And High Delta trades appear to perform poorly on average, although there is a cluster in the 3 month range and Delta 35 that does well. The clusters between the 30 to 60% stop losses and deltas under 22 seem to consistently outperform as well. But again, no very clearcut patterns emerge.

sl-hist-pf

 

sl-diffPF-vs-delta

Expected Returns

Looking at the expected returns, we can see that there appears to be a sweet spot between -30% and -60%, and Deltas less than 22. Similar to what was displayed in the profit factor tables.

sl-hist-expret

We can also see that regardless of the stop loss, high delta trades perform poorly. This is not surprising given the construction of the test.

sl-expRet-delta

Conclusion

Traders may often hold the Front Month trades to expiration but also use stop losses. This test shows that many of those scenarios can be improved marginally, but using a 30% to 50% stop loss on 15 to 20 delta trades seem to gain the most benefit.

The second month, low delta trades with a 30% to 50% stop loss may also be marginally improved.

The tests with Stop Losses between 10% to 20% are significantly impacted across all deltas and all months. These trades are just stopped out to frequently.

Most other tests are only marginally impacted either positive or negatively.

Test Stats

Permutations 270
Profitable Scenarios 174 64%
Total Trades 41402
Profitable Trades 17911 43%
Expired 19394 47%
StopLoss 22008 53%
Test Duration 1.52(min)

Test Files

StopLoss Excel Data

Anatomy of the Backtests

For complete transparency, I want to describe how my tests work at a high level.

Limitations

I am limited by the data I have. I purchased the option data from the CBOE and they only provide end-of-day quotes for all options for every day. I initially purchased 10 years worth of data but found some of their data in 2003 was completely erroneous, as a result, my tests start in 2004 and run through Oct 2012.

I ignore slippage in my tests as I assume that at some point during the day, the premium was higher than at the EOD data (due to theta decay). So if we were to enter a trade early in the day, on average we may be able to actually get filled for the same price as the EOD. Even if I had opening, high and low prices, I’m not sure I could presume to get a better price than EOD. EOD is a moderately conservative approach. An even more conservative would probably be adding slippage of 10-15 cents, but I am looking for general trends here.

Limit Orders/Stop Losses

If a scenario is testing limit orders to sell, if triggered, the test will assume that the order was filled at exactly the limit price. So in the event of gap up/down days that may in reality work in our favor, the test takes a more conservative approach.

Stop losses on the other hand work the in the opposite manner. If a stop loss is in place and is triggered, the test fills the order with the maximum of either the stop loss or the closing price. Again, this is the more conservative approach as we assume gap up/down days affect us negatively.

Delta Tolerance

When searching for conforming spreads, the short strike of the Bull Put or Bear Call spreads is predefined. The test will search for spreads with a short strike equal to or less than the specified strike. However, I also use a tolerance of 10 which limits the range of short strikes that are conforming. For instance, if we are interested in spreads with a delta of 25, the test will return a spread with the largest delta between 15 and 25. This ensures that when there is not too much variety in the spread’s deltas (i.e. close to expiration) that the same spread is not returned for the vast majority of scenarios. This also probably reflects the trader’s mindset. If the trader is interested in a delta of 25, but there is a delta of 29 and the next is 10, they may not find a spread worth taking.

Aggressive Fills

The test is designed to fill the portfolio if there is an open space and there is a conforming trade. As a result, if we have one trade that shuts down very early (i.e. due to IV collapse), another trade for the same expiration may be opened immediately. This may end up with more than 1 trade per month.

Trades Per Month

In order to keep the 1-month, 2-month, and 3-month scenarios comparable, each is designed so that they hold 1 trade at a time per month. For example, with a 1-month trade, only 1 trade may be open at any one time. But a 2-month scenario may have 2 trades with different expirations open at the same time.

Triggers/Adjustments

I can write nearly any type of trigger (i.e. enter or exiting a trade) or trade adjustments so long as it depends on the data I have available. I have the historical option data for the RUT, and historical underlying prices for every instrument. At some point I will write tests that correlated to the performance of the VIX, or put/call ratio of the RUT, etc.