Monday, 3 August 2009

S&P 500 Setup is Bullish for End of August - Sorry!

An attentive reader just pointed out a mistake on this site regarding my S&P 500 setup. I had the signal line for the small traders set at 0.65 standard deviations below the moving average in some places and 0.6 below in others. In fact, the best backtested results came from setting the signal line at 0.6 - as it is on my sample spreadsheets on my DIY page. I had been testing both signal lines a while back, and I guess I neglected to change that on my own spreadsheet! Duh! So I've corrected this now throughout this site.
The main impact is that last Friday's Commitments of Traders data actually means my S&P 500 setup has, indeed, given a bullish signal. This, due to the increase in bearishness of the wrong-way small traders in their derivatives positioning. With the three-week trade delays in that setup, this means my S&P 500 setup will go bullish on the open of trading Monday, Aug. 24. Sorry about the mix-up - and many thanks to this reader (and to an earlier one as well a couple of weeks ago, who wrote in about the same issue, albeit less fruitfully because, at that time, I couldn't figure out what the problem was). Good luck this week!

10 comments:

Anonymous said...

Alex,

Since there's been a tremendous short-term rally in crude recently isn't there going to be increased risk buying Aug 10 ??

Alex Roslin said...

Hi Anonymous,

No, my system doesn't work like that. Besides, I think a longer-term rally in crude would still be in effect whether one buys today or in a week. As well, crude could just as easily sell off again a little between now and next Monday.

Take care,
Alex

In Debt We Trust said...

What is your opinion on gold?

Alex Roslin said...

Hi In Debt...,

Doesn't look good right now from a COT perspective. The setup is in cash. Large spec total open interest - which I'm fading - got excessively bullish last Friday, so that signal has been on sell for two weeks. This past week saw only a small reduction in the total open interest, far from being enough to reverse that signal - as you can see from the latest signals table.

Meanwhile, the large spec net position - now giving a long signal - has risen quite a bit in the past two weeks. It's still far from the signal line to go bearish though. So pretty mixed, with a slightly bearish tilt right now that may or may not mean anything for the near-term.

I'd love to have a silver setup because that COT data can give really interesting results and may work on a different timeframe. On my to-do list!

Regards,
Alex

Alex Roslin said...

Hi again,

I should add, however, for gold that if the large spec net positioning doesn't jump too much next week, the setup will in fact go long due to the time delay for the large spec total OI positioning, which went bullish in mid-June and stayed that way for five weeks, before going to short two Fridays ago.

So that, in fact, is fairly bullish for gold.

It's all because of the varying time delays for my signals. I can't say for sure, however, because the large spec net position signal has zero time delay. So everything depends on whether the large specs get too bullish in the coming weeks.

Regards,
Alex

In Debt We Trust said...

Thanks for the clarification.

Another thing I've been noticing is that the US banks have been repaying TARP. This puts less pressure on the Treasury to borrow. Hence, less inflationary pressure. This could be bearish for gold.

Michael said...

Hi Alex. Thank for very much for sharing your method.

I did my own backtesting using your COT SP setup. I come up with different optimal values and a trade delay of 0. I'm curious how you did your backtesting and could you post some of the statistics such as the profit factor? I'm thinking I must have an error somewhere, if I use your thresholds it's positive but it's not optimal. Perhaps you used different criteria for selecting the best values?

Alex Roslin said...

Hi Michael,

It's very easy to come up with setups that produce better backtested results than those for mine. The question is whether they are the most robust as well. That's really the most important question. Most of the main stats I use are in the backtested results table. Profit-related measures in fact have zero value for choosing my top setups.

If you give me your parameter values, I'd be happy to let you know how those rate in terms of robustness. You might be amazed how badly the most profitable setups do when actually tested rigorously. Also important: did you optimize with raw price data or detrended price data? If with raw data, I would never trade it. That said, you might have found a good one that I missed. I'd be happy to take a look.

If you are interested in more details on proper testing of trading strategies, I just posted more on this in the comments on my DYI spreadsheet page:

http://cotstimer-data.blogspot.com/

Regards,
Alex

Michael said...

I went over my test results and found two things. One was I was inadvertently using STDEVP which changed the results a bit. The other is that your equity curve is almost a straight line whereas the one for the "optimal" values that I found (using a trade delay of 0) was not as flat. It had two big gains and a flat period in the middle. So even though it gained more it is most likely curve fitted around those big gains. So I have more confidence in your chosen values.

When I optimize using a trade delay of 3 I get values that are very close to yours, within 0.05 for your parameters. I get the same value for trade delay, ma period, and stop. So that was interesting that I got the same parameters as you.

I'm going to order the book you mentioned (Evidence-Based Technical Analysis) and try to learn more about your testing methods. Up until this point I've just been running brute force optimizers to find the optimal values and then selecting among the top the ones with the straightest equity curves.

Alex Roslin said...

Good luck Michael! That book took me to a whole new level. A caution: "straight" equity curves are not a reliable way to choose a setup in my opinion. Perhaps you could quantify the "straightness" and then try to validate it against a truly reliable indicator, like Monte Carlo results.

Regards,
Alex