Jeff Augen's StandardDEV study for ToS (thinkscript)

[dbh21]

Might take me a little while. I'm in the middle of retooling my system.

[samerh]

Here's the table with TIF and OVTI added.

 

If you play around with it, the vega-theta is an important aspect of the signal. The IV-HV/LM quotient is still quite variable. At any rate, TIF seems a bad play.

 

 

[Corrected spreadsheet]

Workbook1.xlsx

Edited August 22, 2013 by samerh

[Kim]

I don't think that TIF has ever lost 35% for 5 days holding. Since they usually report BO, you probably counted the earnings date as the last holding date, while it should be a day before.

[samerh]

corrected, thanks for highlighting the error Kim

[Kim]

So how does TIF looks now?

[samerh]

I updated the last post with the corrected spreadsheet.

 

Can't tell on OVTI until NASDAQ gets fixed

 

TIF: vega-theta is 0.05 (so expected IV increase should overcome theta decay), but volatility quotient is 3.8x and so expensive (slight improvement from a few days ago which was at 4.1x but still high).

 

does that ring true with your read of the situation, should we validate this type of thinking?

[dbh21]

Samerh, how are you calculating the IV of the straddle? Are you averaging the call and put IV, weighting it, or something else?

[samerh]

Samerh, how are you calculating the IV of the straddle? Are you averaging the call and put IV, weighting it, or something else?

I'm just taking the average of the call and put IVs. Please let me know if there is a more robust way of doing it.

[dbh21]

Thats the way I am doing it as well. I'm not sure there is a right way. Just trying to align my data with yours. Our IVs are slightly off, and our HV20 calcs are slightly different. 

[dbh21]

Samerh - attached is the output of my backtest for TIF using the SQ (samerh quotient). My numbers come out slightly different for IVs and HVs - and therefor for the SQ. The IV might be due to difference in data (one would hope not). Anyways, the numbers are close. 

 

Can you take a look and see if it looks good. If so, I can run it on a larger data set.

 

-Dustin

TIF-Backtest-SQ.xlsx

[lc_match-aka-Bill]

Do I interpret the spreadsheet correctly if, I see 7 plays on TIF and:

1. I take an average of the "Open Price" of about 5.96 as dollars per  straddle (2 shares actually), or $595 per straddle contract.

2. I take the average of the "P/L" column as 15.9 as the per-contract profit and loss.  (Or $0.16 per straddle)

 

So if I played 5 spreads (neighborhood of $3000 average) I'd average about $75-80 profit before commissions?

 

If so, this is useful information.  Thanks guys.

[dbh21]

Thats correct. Ignore the shares column its an artifact of my tool, but it really represents 1 (shares/100) contract per spread.

 

I'm only using this to check out SamerH's proposition. These are all 5 day spreads which may or may not be the right way to go.

[samerh]

Are you using ATM straddle prices on earnings date or the original straddle strikes from 5 days before? I use the ATM straddle to try and see the effects without gamma gains. I may have used $5 strike spacing in my TIF straddles instead of $2.5. Will check when I'm back at computer

[dbh21]

The straddle prices start 5 days before and are closed the day of earnings. Tell me if I got that wrong.

 

I don't understand the $5 strike spacing on straddles. Are you testing strangles instead?

[samerh]

The straddle prices start 5 days before and are closed the day of earnings. Tell me if I got that wrong.

 

I don't understand the $5 strike spacing on straddles. Are you testing strangles instead?

No I wasn't clear. I think my database used the nearest $5 strike, but TIF has $2.5 separated strikes. So your data is probably more accurate. Here's my TIF numbers re-configured to $2.5 strikes. But even so, my numbers still come out a little differently.

 

 

 

TIF-Backtest-SQ.xlsx

Edited August 26, 2013 by samerh

[tjlocke99]

Thats correct. Ignore the shares column its an artifact of my tool, but it really represents 1 (shares/100) contract per spread.

 

I'm only using this to check out SamerH's proposition. These are all 5 day spreads which may or may not be the right way to go.

Dustin - what is the tool you reference?  Is it a custom database or application you created?

 

Thanks.

[dbh21]

Dustin - what is the tool you reference?  Is it a custom database or application you created?

 

Its a custom backtesting tool I wrote/am writing. The same one I use to produce the earnings charts.

[tjlocke99]

Its a custom backtesting tool I wrote/am writing. The same one I use to produce the earnings charts.

Thanks.  Did you purchase option data as well?  If so then just end of day?  That was always a major reason I never wrote any question software/analytics - getting a-hold of alot of historic options data is expensive.

[dbh21]

Thanks.  Did you purchase option data as well?  If so then just end of day?  That was always a major reason I never wrote any question software/analytics - getting a-hold of alot of historic options data is expensive.

Correct. I bought it. Its just EOD. I don't think I've noticed anywhere selling finer grain data, but that would probably be out of my budget.

[dbh21]

No I wasn't clear. I think my database used the nearest $5 strike, but TIF has $2.5 separated strikes. So your data is probably more accurate. Here's my TIF numbers re-configured to $2.5 strikes. But even so, my numbers still come out a little differently.

 

I think I have my tool ready, but I'm not really sure why our numbers are different. I don't want to report something until I think the numbers are working as expected. Can you expand on the TIF 03/21/2013 trade? What dates did you enter/exit, at what strikes, and for what prices? I get half the P/L you do, so its worth digging into (unless you are including commissions).

        3 

[samerh]

Sure

 

Start date: 3/14/2013

Underlying price: $69.09

Nearest strike: $70

Straddle price: $5.56 (midpoint bid-ask)

Straddle IV: 0.3099 <---- *this is where the discrepancy is. TOS shows this to be 0.3145. need to look into why this is.*

 

End date: 3/21/2013

 

Underlying price: $67.91

Nearest strike: $67.5 (note new strike)

Straddle price: $5.19  (midpoint bid-ask)

 

Straddle P&L: $5.19/5.56-1 = -6.66%

 

 

20 day HV of stock as of 3/14/2013: 0.2138 (Stdev of stock over last 20 days x sort(252) )

Last Move of stock following previous earnings: $63.73 on 11/28/2012 to $59.80 on 11/29/2012 = -6.17%

 

(IV-HV)/LM = (0.3099-0.2138)/0.0617 = 1.56

If I use the TOS number: (0.3145-0.2138)/0.0617 = 1.63

 

 

 

 

 

[dbh21]

Thanks. The closing price is also different - but then we are doing this differently. I follow the original straddle opened on 3/14 and find its closing price. You compare two potentially different straddles. But it looks like everything else is about right, so I will try running this on a larger data set now. I don't think it has to be perfect, just internally consistent.

[samerh]

Agreed. The only advantage to using the different strikes is that I exclude gamma gains in the selection process (which are a bonus).

[Kim]

I think it makes sense to include the gamma gains. I agree that they are a bonus, but they are part of the game plan. We don't rely on them, but over time, a significant percentage of the gains will come from the gamma. it also shows us what is the probability that the stock moves and gives is gamma gains.

 

For example, AMZN will probably result around BE if you look at ATM straddles only - however, taking the gamma gains into account, the straddle/strangle will be profitable most of the time.

[dbh21]

Ok... lets see if this works.

 

Attached is the analysis for the last 2 years on all the stocks in Kim's link. There are 811 trades.

 

I ran simple analysis against 1.7 threshold for SQ (Samerh Quotient), and the results were roughly 50/50. Frankly, if you change the threshold value up or down, the results are still roughly 50/50. 

 

Let me know if I screwed something up

-Dustin

[Gary]

Hey dbh - that pretty much matches what we found earlier... as a trading rule, 2.0 actually worked better than 1.4.

 

If you have some time give 3.5-4.0 a try. A higher threshold in that range seemed like it might filter out some serious losers.

[Kim]

In the file, pass means that SQ >= 1.7 and the straddle was profitable?

[dbh21]

Hey dbh - that pretty much matches what we found earlier... as a trading rule, 2.0 actually worked better than 1.4.

 

If you have some time give 3.5-4.0 a try. A higher threshold in that range seemed like it might filter out some serious losers.

You can tinker with the excel file. Change the threshold to those values and it will update.

 

 

In the file, pass means that SQ >= 1.7 and the straddle was profitable?

 

Correct. The Threshold cell can be updated to what ever SQ value you want and it should update the results.

 

 

Let me know if I had any errors

[samerh]

Dustin,

 

Thanks for all the numbers. I sorted the numbers from low to high on column I "P/L%" and eyeballed the results. I have to say the SQ numbers are all over the place, so while on average lower SQ numbers might be better, on individual trade by trade basis there is so much variability that I'm not sure it really helps. Perhaps its because of theta decay and if we factored that in (where theta decay isn't too aggressive) it might improve, otherwise I'm not sure?

[dbh21]

I have to say the SQ numbers are all over the place, so while on average lower SQ numbers might be better, 

 

I didn't do this before, but lets analyze your equation: (IV-HV)/LM. 

 

In order for SQ to be large, you either need a large numerator or a small denominator (assuming they don't cancel both out).

 

The numerator (IV-HV) implies you want the IV to be larger than the HV - which I think is exactly the opposite of what you really want for straddles. Ideally you want cheap IV and high HV (you buy the straddle for cheap for a stock that jumps around).

 

Similarly, in order for SQ to be large the denominator could be small... meaning the last move was approaching zero. This works against your theory about the market remembering the last move. Large moves, lower the SQ and small moves inflate it.

 

So based on this, you definitely want a smaller SQ. But based on my spreadsheet, you can see if you change the threshold number and reverse the signs in the equations in the Passes and Failed columns, the results are the same. About 50/50.

[samerh]

I agree we want lower SQ numbers, that's what i was saying (I think your spreadsheet was set up to do a buy signal when the sq was larger than hurdle and not smaller) - but it's a moot point as I agree that reversing the signs doesn't change the results much (as we are 50/50)

 

What I meant by "I have to say the SQ numbers are all over the place" was that SQ doesn't seem to be very helpful.

[dbh21]

I think your spreadsheet was set up to do a buy signal when the sq was larger than hurdle and not smaller

Yes. But you can change it by flipping the > sign for a < sign in the Pass and Failed columns.

 

I was just thinking out loud before - but now I wonder why normalize against LM. You have Vol on top and % chg on the bottom. I wonder if it would make more sense to estimate the IV required in order to anticipate the same LM. Something like IV(LM)-IV. If that number is large - it might signal a good buy.