Watch HowardCohodas Trade Index Options Credit Spreads

[HowardCohodas]

Interesting - so did you build a database of prices and historical POT values and options prices?

Can you publish the results here, or are they secret squirrel?

Get a ToS account and there you have it. No secrets. Available to all.

 

[DionysusToast]

So - let me get this straight.

You won't publish any backtest results here & to verify the tests you have done, someone would need to open a ToS account and do it themselves, is that correct?

 

[Martinghoul]

No, I did not implement it within ToS.

The back testing had several different generations as I explored different methods of estimating the probability of the underlying instrument price reaching the short strike. Initially, I used an approach that used historical closing prices of the underlying instrument. With this data I would calculate all the n-day excursions and use the percentage of those excursions as my estimate of the probability of reaching.

There were two or three generations in between. For the last tests I used PoT once I learned of its existence. I saved the end of day data from the ThinkBack application in ToS to include it.

All of this was written in PHP. Sometimes run on my workstation, sometimes run on the server where I host many of my web sites.

Another long answer to a short question.

That's fine, long answers are not an issue. For purposes of clarity and to avoid digression, I am not really interested in any backtests you performed previously which were applied to potentially different strategies. So let's stick with what you refer to as "the last tests".

Correct me if I'm wrong, but, in summary, you performed the backtests externally using end of day data from ToS (PoT and option prices, I would assume)? Which time periods were used for out-sample runs and what was your in-sample period, if I may ask?

 

[HowardCohodas]

Interesting - so did you build a database of prices and historical POT values and options prices?

Can you publish the results here, or are they secret squirrel?

So - let me get this straight.

You won't publish any backtest results here & to verify the tests you have done, someone would need to open a ToS account and do it themselves, is that correct?

You asked me to publish the results of populating the data base. I suggested you could do it yourself.

 

[DionysusToast]

I think that from the explanation - the PoT used in the backtest was calculated by HC & not the actualy POT in ToS. He mentions he calculated "Probability of Reaching".

It is unclear whether he used historical options prices from ToS but if he didn't build his own database of this stuff as he seems to imply, it would be interesting to hear how he managed to do this in PHP which is essentially outside of the ToS platform.

I don't know ToS but I do know that most of these platforms use proprietary database structures for historical price info, so unless they publish the schema definition for the database, it's hard to see how PHP can be used to backest ToS data.

Come on HC - open your box of tricks.

 

[DionysusToast]

You asked me to publish the results of populating the data base. I suggested you could do it yourself.

Very cute Howard.

 

[meanreversion]

Er, at the very minimum could you give us an approximate idea of the backtest output? Any of CAR, max drawdown, Sharpe ratio, and over what timeframe tested?

 

[meanreversion]

I was simply surprised when "back-testing" and "proof" were used in the same sentence in that way.

Your comment is just weird. I asked why you think this strategy should make money. I did not say "show me proof this strategy will definitely make money, using a backtest" as you imply here.

 

[HowardCohodas]

That's fine, long answers are not an issue. For purposes of clarity and to avoid digression, I am not really interested in any backtests you performed previously which were applied to potentially different strategies. So let's stick with what you refer to as "the last tests".

Correct me if I'm wrong, but, in summary, you performed the backtests externally using end of day data from ToS (PoT and option prices, I would assume)? Which time periods were used for out-sample runs and what was your in-sample period, if I may ask?

Yes. I performed the back tests external to ToS with data extracted from ToS. I used data from 2005 through 2009 on RUT and NDX.

Out of sample data was chosen randomly for each run. The tests were done on monthlies so each "data segment" was approximately 60 days to encompass the life of the spread. Thus there were 30 data segments. The number of segments in the in-sample set was also randomly determined in the range of 16 to 24 for each run. The remainder were in the out-of-sample set.

The problem with a fixed in-sample set is that it still lends itself to "curve fitting" tendencies. Randomizing the membership of each set and the set size with each run greatly minimizes this tendency.

Some with whom I've discussed this approach have suggested it is overkill. However, it's easy to accomplish in the software and it increased my confidence that I was not falling victim to curve fitting. This is not the first time I used this technique. Before implementing this approach, I managed to fool myself when evaluating other strategies some years back.

 

[the hare]

Before implementing this approach, I managed to fool myself when evaluating other strategies some years back.

Somehow that does not surprise me.

still, you are to be congratulated on having the good sense to incorporate an element of re sampling in your testing.

 

[HowardCohodas]

I think that from the explanation - the PoT used in the backtest was calculated by HC & not the actualy POT in ToS. He mentions he calculated "Probability of Reaching".

It is unclear whether he used historical options prices from ToS but if he didn't build his own database of this stuff as he seems to imply, it would be interesting to hear how he managed to do this in PHP which is essentially outside of the ToS platform.

I don't know ToS but I do know that most of these platforms use proprietary database structures for historical price info, so unless they publish the schema definition for the database, it's hard to see how PHP can be used to backest ToS data.

Come on HC - open your box of tricks.

For someone who purports to be a high end software guru, your commentary is incredibly ignorant. Furthermore, I told you exactly how I obtained the data.

 

[HowardCohodas]

Somehow that does not surprise me.

Sometimes the lessons I learned were not from the mistakes of others but from blunders of my own making. Such is life.

 

[DionysusToast]

Perhaps you can educate me HC.

In what way is my post showing ignorance? Maybe I can learn something about software from you.

Note though -that you hadn't confirmed you were using extracted data prior to me making that post and you still haven't really confirmed that you have extracted all historical price data from the various strike prices.

I eagerly await you pointing out my ignorance in the ways of software.

 

[HowardCohodas]

Perhaps you can educate me HC.

In what way is my post showing ignorance? Maybe I can learn something about software from you.

Note though -that you hadn't confirmed you were using extracted data prior to me making that post and you still haven't really confirmed that you have extracted all historical price data from the various strike prices.

I eagerly await you pointing out my ignorance in the ways of software.

Oh really? ThinkBack reference in this thread. I made an error in the referenced post. Can you spot it?

 

[Martinghoul]

Yes. I performed the back tests external to ToS with data extracted from ToS. I used data from 2005 through 2009 on RUT and NDX.

Out of sample data was chosen randomly for each run. The tests were done on monthlies so each "data segment" was approximately 60 days to encompass the life of the spread. Thus there were 30 data segments. The number of segments in the in-sample set was also randomly determined in the range of 16 to 24 for each run. The remainder were in the out-of-sample set.

The problem with a fixed in-sample set is that it still lends itself to "curve fitting" tendencies. Randomizing the membership of each set and the set size with each run greatly minimizes this tendency.

Some with whom I've discussed this approach have suggested it is overkill. However, it's easy to accomplish in the software and it increased my confidence that I was not falling victim to curve fitting. This is not the first time I used this technique. Before implementing this approach, I managed to fool myself when evaluating other strategies some years back.

Fantastic, Howard... Looks like a robust methodology indeed.

Now, since I am intending to, effectively, subject your backtest to a critical re-examination, I could do it using the POT values I calculate myself or ones calculated by ToS. Obviously, I don't have an account with ToS and don't really intend to open one. Would you be willing, in order to make our experiment as conclusive as possible, to give me the ToS POT data for a particular time period I require?

 

[HowardCohodas]

Fantastic, Howard... Looks like a robust methodology indeed.

Now, since I am intending to, effectively, subject your backtest to a critical re-examination, I could do it using the POT values I calculate myself or ones calculated by ToS. Obviously, I don't have an account with ToS and don't really intend to open one. Would you be willing, in order to make our experiment as conclusive as possible, to give me the ToS POT data for a particular time period I require?

Absolutely. With all you have done to help me and with the civility you have shown, I could do no less. Besides, I want to know if I messed up.

There is no automated process that I am aware of to extract the data from TOS. I capture and save as a file each day's worth of data into a CSV file. Tell me the time frame you desire, the items you desire in addition to PoT and the index or indexes you would like. With respect to items, I would suggest at least bid, ask, PoT, volume and open interest. One of the items you might also like is probability of expiring. I can provide you with a menu.

If CSV is not your choice of format, I can reformat it to your specifications. Also, the captured files include all series. I could cut down the file size by processing them to remove extraneous data.

 

[meanreversion]

Your methodology for backtesting looks rigourous for sure. Furthermore, in testing from 05-09 you will have encountered pretty much every type of market condition.

What sort of annual return did you get, and what max drawdown?

 

[HowardCohodas]

Your methodology for backtesting looks rigourous for sure. Furthermore, in testing from 05-09 you will have encountered pretty much every type of market condition.

What sort of annual return did you get, and what max drawdown?

I fear they would be misleading and open up an unproductive dialog. The main reason is that the rules I tested are similar to, but not identical to the rules I use in this account. I trade in this account more conservatively than the rules I used in my back tests. Also, some practical adjustment have been made from what I learned in paper-trading. And last, I have added weeklies to my portfolio, and they were not part of the back-tests. I started weeklies in paper-trading.

 

[Pazienza]

I fear they would be misleading and open up an unproductive dialog. The main reason is that the rules I tested are similar to, but not identical to the rules I use in this account. I trade in this account more conservatively than the rules I used in my back tests. Also, some practical adjustment have been made from what I learned in paper-trading. And last, I have added weeklies to my portfolio, and they were not part of the back-tests. I started weeklies in paper-trading.

Genius! "Unproductive dialog" (sic). I bet it would :clap::clap:.

 

[meanreversion]

Wow, you really can't give a straight answer can you.

So you won't reveal anything to us about your backtesting, even though you've done some good work there.

And yet the burden of 'proof' is on us to show your strategy is not going to make money in the long run, and if we can't show this, then it means you're fine.

You sound more like a used car salesman with each post.