Notes for December 30, 2011

[Rembert]

Thanks Paul,

Always interesting to learn new strategies. Connor's TPS strategy might be a good addition to diversify to the rest of the systems I trade. 90% winning trades seems like too good to be true at first sight so I looked up the trading rules. The strategy is one of scaling into a losing initial position by averaging down, a concept that is in general not a good one if you are on the wrong side of a trend. So I was hoping to see some kind of stop loss in the rules to prevent serious account damage in case of a black swan event where the trade goes completely against you. But there is no stop loss. Shouldn't be too hard to implement one of course. Do you have any toughts on that ?

Is there a site that gives signals for Connor's TPS strategy or is it something one has to implement on their own ?

Also best wishes to all for 2012, may it be a trending year.

update : found a recent Larry Connors webinar and some discussion on the use of stop loss orders with his strategies http://forums.worden.com/default.aspx?g=posts&t=53070

[grems8544]

Connor's TPS strategy might be a good addition to diversify to the rest of the systems I trade. 90% winning trades seems like too good to be true at first sight so I looked up the trading rules. The strategy is one of scaling into a losing initial position by averaging down, a concept that is in general not a good one if you are on the wrong side of a trend. (bold added by PGD for emphasis)

Absolutely agree 100%. I got my rear handed to me this year, losing almost 10% net, when I was fully engaged in the TPS strategy in multiple accounts and we went down, and kept going down. This is a weakness of the Connor's system that he does not mention in any of his literature (of course not, it wouldn't sell), and it's proof-positive that the strategies need to be tweaked. More on this below after I address your other comments.

So I was hoping to see some kind of stop loss in the rules to prevent serious account damage in case of a black swan event where the trade goes completely against you. But there is no stop loss. Shouldn't be too hard to implement one of course. Do you have any toughts on that ?

I do, and I'll address that topic separately, as I've done considerable work looking at Stop Losses and his strategies. Note the MAE column tells you about your intra-day drawdown, which (psychologically), will give you a taste of the main course. In general, SL significantly reduce the profit factor (PF) as well as the W/L ratio, so the better path is to not have entered the trade at all. This is accomplished using a macro signal outside of the Connor's strategy. Again, let me get my material together in a way I can present it coherently and I'll address this important topic.

Is there a site that gives signals for Connor's TPS strategy or is it something one has to implement on their own ?

HGSI implements a filter and every night the candidates are available.

Hsin Yen has written some TradeStation code for the radar screen function (indicator) and I've modified it for my use as well as have implemented management strategies using his foundation. Once in a trade, I use the strategy code to manage the trade so that I do not have to be present at 3:59 pm to execute the sell. This is a deviation from the rules in the book, but I find the statistics somewhat improved.

Chris White has written ETF Bandit, which is a stand-alone program for scanning Yahoo! data (end of day is base package, intra-day is an add-on that is totally worth it). Here's the link: http://www.edgerater.com/products/etftradingbandit/ Tell Chris I sent you -- I get nothing for the referral except an exchange of good will.

Also best wishes to all for 2012, may it be a trending year.

Gosh, I hope so. I'm tired of treading water. Same to you and all of us here.

==================

So, onto the topic of "When Connor's TPS Fails", or better entitled "How to Have Your Butt Handed to You when Reality doesn't Match Backtesting"

Take all of the Connor's TPS trades on the long side since the beginning of time (or available data) and trade whatever is available. The equity curve looks like this:



Good news: lower left to upper right -- Over a long time, it appears you'll make money.
Bad news: large volatility causes spikes, and this (psychologically) will kill your ability to consistently trade the strategy. It will also hurt your bank account, as it did to me this past year (note the "Yeouch" highlight).

The statistics look solid too:



Worse case drawdown on the long side was 20%, and worse running losing streak was 8 consecutive losses.

What keys you in that this strategy has legs is that the distribution of returns is positive, specifically the mean and the median:



Despite the positive bias, note that the tails to the left and the bars below 0 are significant, and if you're heavily invested in those ETFs when they lock in their losses, you're losing money quickly.

So the goal here is to minimize losses, so that the distribution improves on the right side of the graph above and that the number of trades below 0 decreases.

While obvious to me now, it was not obvious in my original backtesting, simply because I was not looking for failure mechanisms, is that THE LENGTH OF THE TRADE IN DAYS has a material impact on P/L values. Here's some findings:

This is a snapshot of the trade blog, sorted from highest wins downward. The trades are constructed differently than I've presented in other places -- here, instead of a fixed ceiling of (say) $6250, I am buying 100 shares on entry, then 200 shares as price drops, then 300 shares as it drops further, then 400 shares. A fully loaded-trade would be 1000 shares of the equity. This is the strategy of the TPS system.

Here's the trade blog, with the best performing trades at the top:



Note that:

1) in general, the most winning positions do not "fill" to the 1000-share level before they are executed for sale.
2) in general, the number of days to achieve a winning position is single-digits with rare exception.

Also note that with a little mental math, that the P/L % value divided by the MAE% value is often quite larger than 1.0, which means that you've taken less risk for achieved reward.

Hence, we can conclude from this presentation that higher probability trades typically do not "fill" to the 4th buy (e.g., we're holding 1000 shares on the 100-200-300-400 TPS buy schedule), and that higher probability trades last relatively short duration in time (e.g., single digit days).

As further example of this, take a look at the same listing of trades, but now with the worse trades at the top:



First of all, note the following:

1) many of the bad trades fill to 1000 shares (all four scale-ins are executed)
2) The number of days to the exit point is more-than-often double-digit, e.g., greater than 9 days
3) The MAE% is incredibly poor, e.g., volatility is higher, intraday drawdowns are higher than normal, and overall, the reward/risk ratios are quite poor (P/L% divided by MAE%).

Hence, we can conclude from this presentation that we need to pay attention when we have fully "scaled in", as this fact, in combination with the hold time of the trade, can give us indication that the trade is a candidate for failure.

So the question now becomes "How many days is too many to hold a position?" and "Should we have another rule that states that after X days we sell no matter what?"

To help answer those questions, the following chart was created:



Here, I've rescaled between +20% Profit and -15% Loss to better visualize the data.

What this chart shows is that as the duration of the trade marches on, the probability of failure increases. Note that in the entire history of trades that durations of less than 4 days have NOT resulted in losses. Durations of 5 days to 9 days typically result in positive trades (but not exclusively), as indicated by the trend line, and certainly past 10 days (where the trend line crosses 0%) the odds are moving against us.

Correspondingly, trades lasting longer than 9 days should be exited from a historical view point, and trades longer than 4 days but shorter than 10 need to be watched for a forced exit.

As I alluded to above, the key here is to identify when not to get into a trade, rather than trying to figure out what to do once the trade goes against you. Also, Stop Loss settings will trigger according to the the MAE% criteria, effectively locking in the losses, so in general, Stop Loss settings should be avoided with the Connor's TPS strategies. More on this latter point in a different post.

Regards,

pgd

[lisa]

paul,

maybe i'm being a maroon, but in the TPS test results and comparing the chart of QQQ to those first couple of best-performing trades from the trade blog, they don't seem to match up. the 3/13-3/20/00 trade looks like it should have been a loser, but it made 144% in only 7 days. even considering the internet bubble, 147% seems a little outrageous in such a short time. the next couple of trades looked similarly unrealistic. what am i missing?

i'm sure you've already pondered this, but i started looking at them and wondering if you flipped your entry rules since the bigger gains seemed to be made early--so buy 400 first, then 300, 200, 100--and add a rule of not holding more than X number of days.

thanks,
lisa

[grems8544]

paul,

maybe i'm being a maroon, but in the TPS test results and comparing the chart of QQQ to those first couple of best-performing trades from the trade blog, they don't seem to match up. the 3/13-3/20/00 trade looks like it should have been a loser, but it made 144% in only 7 days. even considering the internet bubble, 147% seems a little outrageous in such a short time. the next couple of trades looked similarly unrealistic. what am i missing?

i'm sure you've already pondered this, but i started looking at them and wondering if you flipped your entry rules since the bigger gains seemed to be made early--so buy 400 first, then 300, 200, 100--and add a rule of not holding more than X number of days.

thanks,
lisa

Hi Lisa,

No, I did not flip the entry rules. You can see this for yourself -- in the column "Tr.", which lists the number of actual trades for that position (knowing that a full position is 1000 shares, column "Sh.") -- we see that when Tr=1 that Sh=100, and when Tr = 2 we have Sh=300 (100 + 200), etc.

I agree with you that something seems a bit fishy here. The data source that I'm using is HGSI with a MetaStock export. It seems there is a major discontinuity of the data which makes anything before 2/3/03, and indeed, 200 trading days past this date, suspect. Here's HGSI's view of the data:



EdgeRater, the software I use to backtest strategies, allows the selection of different data sources. A review of Yahoo! historical data for QQQ (attached here: QQQ-Historical-Yahoo.xlsx) shows very different pricing for the Q's, and more importantly, no discontinuity, so obviously, the MetaStock export is in error. I'll send a note to HGSI and we'll see if they can figure out the issue.

As far as I can tell, trades from 11/17/2003 (200d + 2/1/2003) onward are completely valid when comparing HGSI/MetaStock data and Yahoo! data in terms of prices. Considering data from 11/17/03 removes 377 trades of the previous data set, leaving 1120 trades to consider. The previous results are still valid -- trades longer than 9 days in length have a greater degree of failure associated with them than do the trades with shorter time frames. Further, MAE, which is the number which will cause us to panic intratrade, has nearly the same average and standard deviation from 11/17/03 onward as it does for the entire history of the Q's, which is reassuring.

My take away here is still the same -- the goal is obviously to not experience the horrific drawdowns possible with the strategy, and the real question becomes how to prevent entering the trade when conditions are such that we may be in trouble in the bigger picture. I'm presently looking at the status of GGT from 9/08 onward to better understand the setups of when Connor's fails in context of overall market conditions.

Keep the questions flowing ... enjoy the digging.

Regards,

pgd