Chasing a rainbow?
Author: warren631
Creation Date: 1/13/2009 1:39 PM
profile picture

warren631

#1
I'm getting very disillusioned with trying to find the 'holy grail' of technical trading systems. I have tried my own and all published scripts for over three years, tried various versions of Wealth-Lab and custom VB programs, without finding any consistently successful trading system.

I have found many which, with optimization, work very well on historic data. But when I use them in real-time automated trading they all fail.

Are technical indicators just illusionary? Oscillators are low at good buy times and high at good sell times. But are they just replicating the data without any predictive capacity? Lows stay low or go lower, highs stay high or go higher.

I am not asking for details, I just want to know if anyone has really had any real concrete consistent success (consistently better than B&H) trading on technical indicators in real time - or am I just wasting my time chasing a mathematically impossible rainbow?
profile picture

DrKoch

#2
QUOTE:
I have tried my own and all published scripts for over three years, tried various versions...

I did the same, starting in 2001, worked on this nearly full time, nearly all the time.
I had very similar findings and I am pretty sure that most TA Indicators are worthless.

Things changed when I took two things into account:

1. My own indicators. Based on quantitative analysis, random walk theory, statistics, physics, ...

2. Good (intelligent) data mining. Using advanced statistical methods to find "predictive power" from massive combinations of "Indicators".

The combination of these two items results in working, profitable trading systems.
But: you'll need software way more powerful than WL to implement it.
profile picture

Roland

#3

We’ve all tried most of the scripts, analysed the procedures and evaluated performance over various time horizons with short or long watchlists. Some of the scripts were downright ridiculous in the sense that they could not, even over a century, produce a positive result while most could not even beat the long term Buy & Hold strategy.

In this perspective, Warren is right. You get disillusioned when ever you try auto-trading live. It is then that you find that the best well-laid plans fall apart and perform miserably. But this is understandable; everyone wants to have a simple script that can handle every situation. And that is unreasonable. How could a single simple script handle the diversity of stocks that it should cater to? How could a simple solution tackle such a complex problem where every single stock has its own signature?

René is right in his analysis and there is even more. If you look closely at Stochastic Portfolio Theory you will quickly find that beating the secular market average return is relatively hard to do.

It is not short term that you need to beat the averages; it is long term. It becomes irrelevant if on short periods of time you achieve positive return if when looking long term you blow up your account. It does not matter if you win or loose a battle; it’s the outcome of the war, your broker statement that matters. Curve fitting and over optimization are part of self-illusion and a major failure reason to many wannabe trader.

Be sure before you go live that your trading procedures do in fact perform positively going forward in time. This means that your procedures should account for the unknown, the unforeseeable in order to protect your portfolio and your capital.

profile picture

warren631

#4
I'm glad to see I am not alone in my quest while others are making millions. I have only been trading an IB simulated account to force myself to first prove my skills in a real-time environment before risking capital - and I am very pleased that I made that decision.

I have been focusing on technical indicators on 1-minute charts of QQQQ. The random noise at that level makes it impossible to see the forest from the trees. So I am moving to daily charts.

I also believe indicators can only provide limited assistance. I must also watch the chart patterns, support/resistance, etc.

I must also find a system that I am completely confident in so that I can psychologically better withstand the inevitable draw downs.
profile picture

MikeCaron

#5
I have also had my hopes destroyed when moving scripts from the lab into real time use. I am still searching for the set of scripts that would be helpful.

Some thoughts I have wondered about are:

The VectorVest approach is to apply the hottest script of the week until it is clearly out of favor. Anyone that has seen this tool knows that they rank their scripts on a weekly basis and make recommendations on what to follow. They also have a proprietary indicator indicating when it would be good to enter the market. Categorizing scripts into some generally market conditions would help with figuring out when to apply them (trend following, market reversions to the mean, high volatility).

Regarding time frames, I was thinking that 15-minutes would be better than 1 minute in terms of less noise and allow for better position entry than EOD. Any opinions on this one? I am currently working on EOD strategies and then hoping to refine them.

What instruments will be used for the analysis? Stocks, ETFs, eminis and currencies come to mind.

Regarding Dr. Koch's suggestion, I have had similar thoughts but limited action. Statistically analyzing the myriad of indicators is doable using one instrument and timeframe, I think. However each indicator generally can operate over a range of time periods (SMA(1), SMA(2), etc.). The strategy can also operate over a set of time periods (1-minute, 5-minute, 1 hour, EOD). How does one get started to build this kind of analysis? I installed R, found a technical indicator library, ordered three volumes on statistics:
- "The R Book" (more of a tome)
- "Statistics and Finance, An introduction" -- the introduction was very helpful, and
- "Applied Econometrics with R" (not apparently useful at this point).

I included a few primitive indicators at some standard time intervals, (RSI, StochRSI, EMA(3,5,8,20,50,200), Pivot Points (daily, weekly), Fibonacci retracement and projection levels, Volume, ROC, momentum) plus the zigzag indicator (as the reference for ideal performance) used SPY (an approximation of the S&P emini), fifteen minute time interval, and generated a lot of data. I loaded this data into R, read the primer in Statistics and Finance, and figured out that I need to focus on time interval, multi-variable analysis. I tried to do a pairs analysis to find items that would correlate to the ZIgZag indicator. However, because of my lack of skills, I did not get far. Any suggestions on how to dissect this data? Is this a similar approach that Dr. Koch pursued?

Another thought I had in mind was capturing the rules used by a chart technician, but finding someone that is successful and can clearly state what they do and even would talk me has been impossible. I’d probably have better luck analyzing their trading records with this approach. I suspect the rules are very complicated and they adapt them all of the time to different market conditions.

On the chart technician strategy, I also read the “High Probability Trading Strategies” book by Robert Miner and that looked promising. He has his own tool for doing the trades, though he says that some parts could be automated and there are some parts that could not be automated in his book. Anyways, from a career perspective, this looked like it could be promising though not completely automated.

So, like Warren631, I am at a cross roads as well.
profile picture

DrKoch

#6
Mike,
a few remarks:
I think its not useful to focus on a single instrument. I always use groups of instruments (like 100 Naz stocks or a few sector indices)

The key is to generate a big set of data points to give the algorithms some "food".

Hint: It does not help to go too far back in history (in order to get enough events) because things that worked 10 years ago don't work today...

You can't use most standard indicators directly. An EMA for example moves with price and is different for each instrument, month, etc. You need "scaled" values like Rate-Of-Change in percent or EMA(10) rel. EMA(20) in percent and so on.

I also use "R". (for cone: That's what I referred to with "more powerful")