Friday 12 December 2008

New SPX Setup Bearish

Lots of interesting new signals from this afternoon's Commitments of Traders report. I've updated my latest signals table here. My brand spanking new S&P 500 trading setup (see more on that below) has gone to bearish for next week's open. (It was in cash last week.) The Nikkei has gone to cash, as has natural gas. And platinum is now bullish. All for next week's open of trading.

That table also has the backtesting results of my new S&P 500 setup. The specs are also all posted in the notes to the table. It's the product of a bunch of new testing I've been doing using detrended market data (data that has the long-term bullish or bearish bias of the market stripped out) and comparing it to randomly generated price data using something called a Monte Carlo test (invented during development of the atomic bomb). I plan to stick with this new setup until my next round of updating all the setups in mid-2009. I will also now resume trading this particular setup with no added technical indicators for backup. Sorry for switching this one around so much in the past few weeks, but I think the effort and your patience have paid off. You'll also see on that table I've added some new metrics that I'm using to evaluate the best setups. I've just published a spreadsheet for the setup on my DYI page. Have a good weekend!

TAGS: S&P 500, SPX, Nikkei, natural gas, platinum, COT, Commitments of Traders, derivatives, Black Swans, market timing, trading system development, CFTC, Commodity Futures Trading Commission, COTs Timer, out-of-sample testing, walk-around testing

16 comments:

Unknown said...

Alex,

I'm interested and will review the signals more carefully... but... I am unclear on why you delay the trade from the set up (i.e., the data).

I trade futures and options and while you may be able to see some "extreme bullishness" in the aggregate (negating the need for any one investor to actually be holding for the period of delay), I am as likely to be on the opposite side of a trade minutes/weeks/days from now as a coin toss. But am I wrong about that very fact?

Is your thesis that if traders are in the aggregate in an extreme position, there is some "psychological persistence" that will affect the markets deep into the future (hence the delay)?

Or is there some simpler explanation that I'm just too thick to appreciate?

Thanks,
Rick

Unknown said...

Alex,

On your Latest Signals column, does it take into account the trading lag time? SO is the SPX went bearish, does that mean you will trade bearish in 3 weeks? Or is the current bearish call the result of waiting for 3 weeks already?
Also, what if your signal goes bearish, stays there for 2 weeks and then goes bullish - does that negate the trade signal or do you always take the trade signal plus 3 weeks?

Thanks much.
T

Alex Roslin said...

Hi T,

Thanks for your questions. The simplest way I thought to do it is to sync the two signals on the latest signals table so it just shows the next-week signal. But if you check the spreadsheet, you'll see the formulas for each of the two signals reveal that week's signal, without any lag. The combined signal includes the lag.

Also, a signal in a following week doesn't negate the previous week's signal.

Regards,
Alex

Alex Roslin said...

Hi Rick,

Thanks for your message. There's not necessarily a need to delay the trade. I just found there was a reliable improvement in backtested results when I used a delay in some markets. Not in all, mind you. I'm not sure why this market effect would exist. I figure one reason is because it can take time for positioning by larger players to have an actual market impact.

Regards,
Alex

Unknown said...

Thanks for the response Alex.

SO, am I to understand that you do not make public the std dev data for the other markets? I would understand if you did not but I just want to be sure so I do not continue to hunt on the site. I believe I have looked through every comment and response and post at least twice.

Also, personally, what has your performance been since you started investing real dollars in 2007?
I ask this b/c I ran a quant fund that backtesting showed enormous returns but we could never duplicate them due to trading costs and entry points and exit points.

Thanks,
T

Alex Roslin said...

Hi again T,

That's incorrect. The parameter values for several other setups are available in my latest signals table notes - gold, NDX and the 10-year Treasury. But I would caution that I am refining all of those using detrended data and Monte Carlo testing, so I'm not trading any of those setups right now without combining them with technical indicators.

Regarding my own returns, I've revised my setups so many times as I've developed the setups that it's not relevant. As I mentioned, I've just started combining technical indicators for most setups because I learned of new steps that are quite important in testing.

Without knowing more details on your quant fund backtesting, it's hard for me to comment. Did your quant fund use detrended price data and Monte Carlo testing for the backtesting? What were the backtested results? What were your risk-control methods? How long did you trade it in real time before giving up?

Regards,
Alex

Unknown said...

Alex,

1. The quant fund was run for 3 years with client money. I decided to shut it down because the market data was showing us (back in May of 08) that the data was outside of a historical normal distribution. We backtested it back for 70 years of data.
Simply put, the volatility in the overall market negated individual equities moves. Rising tide lifts and recedes all ships is magnified during times of great overall market vol. Stand outs cannot stand out - long or short. All in all, out investors lost 5% in that particular strategy. I only count absolute performance - success is when you have more money than before - not that you lost less money than the index.

I understand the performance issue. You cannot state that you performed X% if the underlying metrics have changed - not duplicable for future results.

I just found the notes section and have read them all.

Since only 2 of the strategies have Monte Carlo scores above 95, am I to assume that the others should not be trusted as of yet - that the returns may be due to randomness chance?

T

Alex Roslin said...

Hi T,

Thanks again for your message. Just noticed I hadn't updated that latest signals table correctly. Only one setup so far has full results, including Monte Carlo testing. So that's the only one I am trading right now without use of technicals.

I also noticed an increasing number of stops getting hit with my setups through the summer. The stops are based on the past largest drawdown, so what they were saying was that markets were outside historic norms, as you found too. However, my COT data goes back only to 1995, not 70 years. So your data is quite interesting and reflects the gravity of what's going on.

In recognition of this problem - and the fact that markets don't typically reflect a normal distribution of data - I've decided to use a couple of risk-control methods: my stop at the largest past drawdown in backtesting and my Black Swan Rule, which says that if the portfolio loses 6% in any four-week period I stop trading for four weeks. The idea is to give the data time to readjust itself. Remains to be seen if that's enough time, but the rule would have kept me out of much of the recent volatility.

One observation is about your backtested period. Seventy years of data gives good statistical robustness but may also give less responsiveness to current conditions. There is a tradeoff between the two issues and some debate as to the optimal backtested period. Did you address this in your system?

Regards,
Alex

Unknown said...

Alex,

Our backtesting was used to provide a baseline of performance. The problem with so much data was that the larger the sample size, the more likely reversion to the mean occurs. So, what we did was take the historical data and the overlay shorter time periods with volatility data - which is difficult since volatility measures are not 70 years old.

What we saw was that as short term volatilty increased, we had to also decrease our holding periods and tighten our stops. Most people increase their stops when volatility increases but that just leads to death by 1000 cuts.
It the market speeds up to 90mph (or kmh where you are) then the "speed" of the portfolio must also increase. Also, the new "speed" became our baseline from which we then measured forward moves.

On your 6% stop, has then been arrived at by using position sizing and volatility metrics? I have found that hard stops that are not market weighted can sometimes provide a false sense of security.

Generally, in my naive opinion, I believe the way you are using COT data is actually quite brilliant. You are looking at it relative to the most recent market conditions; almost in a rough way measuring the "acceleration" of the COT data.

I would say that the Monte Carlo tests are vital because in my opinion, a system is only good if it also can find correlative market conditions to execute in. No system works all times so my search is for those metrics that will work in specific market conditions. Simply put, if the system only works when vols are between x and y and etc, then that is the only time I will use that system. Market. In my simple mind, a system looks for identifiable market conditions.

Thanks for your kind responses by the way.

Alex Roslin said...

Hi T,

My 6% portfolio stop is just a twist on a common trader rule that's mentioned in Curtis Faith's Way of the Turtle. I haven't tested it myself.

Fyi, as an idea, I've been meaning to test at some point the DeMark system, including trades based on TDST lines. Might be worth looking into.

Regards,
Alex

Unknown said...

Alex,

I have read the Turtle books. My understanding is that their system has underperformed since about 2005. Further evidence that certain systems work only when certain market conditions appear, perhaps.

For my position sizing, I use a risk weighted formula.
Roughly (I have tried to adjust it for shares instead of options)


I buy SPY at 100
My stop is 90 - 10% down
My total portfolio is $100k
My max risk is 3% (a number from 0 to max 5%) or $3000
Divide 3% by 10% or $3000/10%=$30,000

So, I initiate SPY at $100 so I can buy $30,000 / $100 = 300 shares of SPY.

My max risk for the overall portfolio is a function of the win/loss ratio and the probability of winning as well as the return per win/loss position. I also use expectancy calcs.

Now, I have at least 5 to 10 positions at any one time, across 3-4 strategies. Tharp has commented on the importance of having enough positions to make trading worthwhile. If you have a 65% chance of winning at blackjack, the more you play, the more you win, and the preformance will revert to the mean, which is the % of wins.

This sizing formula will not work if you only have 2-4 positions as the win amounts will not affect the total portfolio enough. I use a slightly different formula for my options trades.

Sorry for the long winded emails.
Today the market is like watching paint dry so I have some time to post and talk to clients.
T

Alex Roslin said...

Hi again T,

Thanks for your message. All very interesting. What are your entry and exit criteria based on?

Alex

Unknown said...

Alex,

1. On ETF's, I use 2 things: technical support levels as well as the underlying option contract's volatility. But see, I trade for ranges inside of range while you are looking for moves beyond the normal distribution.
2. On options, I have a formula that I use that combines volatility with probability with standard deviations with technical support resistance levels. It's not as complex as it might seem. I simply look for those underlyings that have support /resistance levels OUTSIDE of a 1 std dev move. 1 std dev because that gives me a roughly 66% chance of success. I do fine tune the std dev with the volatility metric though.

So, simply, I am the house at vegas. the odds are with me that over time, I will win 66% of the time. I restrict my winners with the previously mentioned position sizing and as long as my average win is about 25%, I can significantly outperform the market and have defined risk.

I am interested in adding your strategy as a hedge tool as well as a "homerun" tool.

I am inherently boring and could not care which direction the market goes. I want to make money up or down and typically do.

Thanks
T

Unknown said...

Alex,
I forgot - my exit criteria is based upon profit rules as governed by time and volatility as well as technical s/r levels.

T

Unknown said...

Alex,

On your COT Timer sheet, am I reading the formula correctly that if the 3 weeks prior it is a buy and the current week is a buy, then you buy but if they are both sell then you sell and if they do not confirm it is cash? If so, is this what you have backtested? Your verbal descriptions are that you take a signal and add 3 weeks. That could be done on excel with a simple date function instead of the conditional formula you have there. I just want to make sure as I program mine, that I using formula's consistent with what you backtested.

Thanks!
T

Alex Roslin said...

Hi T,

That spreadsheet is what I have tested for SPX. Setup one has a trade delay of two weeks, setup two of three weeks. I tested it and all the other possible setups on detrended price data, but the results on the latest signals table are for the raw price data.

Regards,
Alex