Does anyone know of a model using moving averages, RSIs, or stochastics on the broad market to help identify when the market is likely to drop more than 3%? Thank you.
First, I developed a historical indicator model, which I published at GT in 2002, that portends relatively large magnitude moves on the Spx. I used the model as a confirmation set for the Cronus model, which portends large magnitude moves. The Spx declined 10% in 4 trading days from the date of the published short on 04.17.2002. Later, I extended the model to the hourly charts. Second, Cronus, which is a deterministic, astrophysical measure, was developed specifically to depict large magnitude moves on multiple time frames, and I published a short on the Spx on 04.29.2011. I can provide a copyrighted copy to qualified parties. What is the nature of your interest? John
Check subsequent index volatility (for 1, 5, or 20 days) from any close where a short moving-average of closing prices is above or below a long moving-average of closing prices. I suggest you start with the time-honored ‘Golden Cross’ of 50-day and 200-day averages and then step the parameters around to see how robust the results are (short MAs of 20-100 days and long MAs of 100-300 days).
You could also clearly see that the VIX of the S&P500 (or the S&P100 if you check data before 1990) will be different for the index when it’s trading above/below the moving average crossovers you test. I mean the VIX will be, statistically speaking, provably of a different distribution.
Index behavior is heteroskedastic, but conditional to ‘trend’. Broadly defined, ‘bull’ markets are characterized by lower vol than ‘bear’ markets.
I’m certain that if you defined an algorithm using an MA crossover and some measure of volatility (actual recent realized or options-related e.g. VIX), and maybe some recent return info (ROC over last 20 sessions?), you’d have a pretty good logit model of when the index is likely to drop X% in the next N sessions.
Having built that – the question is: How would you USE it?
, First, Bull markets necessarily being characterized by low vol and vice-versa is not a true statement, and I am presuming that you are referring to equity markets too. Second, markets can experience increasing prices for extremal events too. Look at the first to mid part of February 2011 for the S&P500 as a counter example for the vol statement and an example of a positive extremal event.
you don’t seem to understand phrases like ‘broadly defined … characterized by’ or the concept of ‘general rules’. Let me explain: counterexamples don’t disprove general tendencies, general tendencies are defined statistically by large numbers of data points.
Put another way, you can get occasional bad results when betting on good hands, but in the long run, if you play the odds, you tend to win more money. That’s a general tendency with a set of counterexamples. 🙂
You can see by my previous post that I’m writing about equity indices exclusively, as I assume the original poster was asking about equity indices, given his reference to the ‘broad market’.
So while we have periods of high VIX or examples of volatile moves when the equity index is an uptrend, the fact is that periods of index volatility cluster during periods of downtrending equity indices.
Anybody with five-to-ten minutes, a spreadsheet, and access to Yahoo!Finance can dowload the VIX, the S&P500, calculate the MAs, then measure the StDevs of returns and view the distribution of VIX both in/out of trend, and view for themselves the simple truth that vol is trend-dependent.
Best of luck with the astrophysics!
Thank you for your valuable time. When doing proofs, only one counter example is needed to disprove a general statement, although I will present another in a moment. Thank you for your education on statistical analysis. I had taken some Phd level classes at GT, which included multi-variate statistical analysis, so I will try to keep up with your explanations. Thank you for the education on money management, although I did write a thesis on money management that incorporated Ralph Vince’s Optimal f.
First, I am presuming that you have performed runs test that support your statement on multiple time frames with p-value less than 5% to support your negative one correlation statement, although I would bet that is not the case, and I am not a betting man. Bill, I am not here to pick on you because most folks make the same statement that the IV increases as prices go down. Second, a distributional test will not necessarily reveal the significance level of the correlation between the series. I am referring to your statement comparing the distributions, as a test of correlations. Counter example #2……The Vix reached a high of 19.07 on 4.18.2011 with the Spx low at 1295.58, and the Vix reached a high of 19.29 on 5.6.2011 with the Spx at 1335.58. Clearly, the Vix is higher on a daily basis with the Spx 40 points higher. Counter example #1…..the Vix closed at 15.93 with the Spx close at 1310.87 on 2.4.2011, and the Vix closed at 16.43 with the Spx close at 1343.01 on 2.18.2011. Clearly, the Vix was at a higher level with the Spx at a higher level in both cases. I could go on ad nauseum with counter examples before even starting the appropriate test for correlations. Also, I posted on Twitter a Spx long@1298 at 12:32 EST on 4.18.2011, and I have a copyright of a Spx daily short for 04.29.2011 with accompanying Cronus measures through May 2011. Note, I did not say that the daily short would last throughout May 2011. For qualified individuals, I will forward the copyrighted copy to support my statements. Folks, thank your for your time
Difficult to find a model that turns that fast without getting in and out so much during the trend it’s easy to end up losing in a nice uptrend. Certainly you can try some filters like only happens after a certain amount of previous uptrend or a range expansion and a million other ideas. I had some success with key reversals under some conditions. So I think it can be done just not every 3% move.
all the luck in the world in his quest for a trading signal possessing 100% win rate. Apparently he has provable hits twice in 10 years with his model. We mere mortals will have to labor on with our general tendencies, since trades that come around twice in 10 years make for rather dull intervals, assuming zero losers….
rmchair theorists aside, any member of this group with a decent model for such is not going to just disclose it in a free and open forum.
Back in the 90’s I sold some of my models and trained fund managers and traders, but I was paid handsomely. This I did in addition to running my own funds and was like writing calls against my models.
In the 80’s and 90’s I also used to buy models and was an active member of Club 3000 (a forum for discussing technical models).
One of the platforms for selling trading systems is still around although the founder John Hill has gone off to set up his own funds… Gibbons Burke also used to be there.
Bob Pardo did the same, but he only sold his own systems and they were crap. Larry Williams models left Bob’s in the dust. All these models are relatively simple blunt instruments, but some have proven the test of time.
Here are some pointers though…
As inferred by Bill above, Financials tend to panic down, whereas Commodities panic up. FX can go either way… but this isn’t the key issue as John indicated. None of my current models have directional bias (why is a whole other discussion though).
Equity indices as an amalgam of diverse underlying instruments tend to trend with greater volatility than other more homogeneous markets. For this reason (and the contract sizes) I have never traded equity indices on a time frame longer than 3 days (I generally day trade).
For this reason I have generally found moving averages inappropriate for equity indices as they lag too much. Even methods such the Turtle style channel breakout does not work well in the equity indices (or a combination of both), despite working in Bonds et. al.
As for overbought/oversold indicators like the RSI, the problem there is over-fitting. If you are only looking at single time frame data I do not believe you will be able to find a statistically robust model using overbought/oversold indicators…
To spell it out, if you want to use overbought/oversold indicators you will have to work with multiple time frames, otherwise I suggest you move your investigation to short-term volatility breakout… or move into the more complex areas of, cycles, waves, harmonics and chart patterns, but as John will tell you, this can be a very, very long and arduous path before you get ANY reasonable results.
I’m sure John will confess it is worth it in the long run. But very few traders I find have the tenacity and attention span to pursue this path, which of course means there are still good profits for those on this path… As for the basic channel, volatility and moving average models… there are a lot of people using those methods reducing profit any new player can pull out using them.
P.S. If you want to rent some black boxes you can have a look here…
I now post my TRADING ALERTS
into my personal FACEBOOK ACCOUNT
. Don't worry as I don't post stupid cat videos or what I eat!