This only means that 90% of indicators are actually only variations of 5% of those that don't work (let alone the very fact that an indicator cannot "beat" the market without a strategy).

1. Indicators / calculations on the price (averages, oscillators, exotic bars, etc...) are always lagging... It is curious to see how people are trying to find the holy grail in the wrong sweet spot ;-)

2. In 20 years time the markets (and their dynamics) have materially changed. In the 90's it costed monthly 1.000$ to have a 64Kbs leased line, not enough to have a decent real-time feed. Today monthly 120$ will buy you 100Mbs cable and redundant 40Mbs VDSL2 backup.

3. Today 10K$ will buy you a state-of-the art machine, suitable for smart trading with HD 6 screen panoramic view and able to run parallel complex algo's with real-time neural network scoring...

For me the best test remains the forward test, next week and the week after and the P&L on my account...

I'm also missing market 'context'. Different market required different approaches.

In my opinion, there is no silver bullet over a 20 years time span...

Maybe it was unclear. 5% of the 766 technical analysis variations we tested beat the SPY total return for the 20 year period.

Unfortunately it's not clear still. What is a "technical analysis variation"? Imagine that you explore a classical indicator, say, RSI. Then will "RSI > 70" and "RSI > 69" different "technical analysis variations"? If so, I don't understand why you came up with such a small amount of variations. Please clarify.

Would you please tell us which tech indictor perform better than others? This is extreme beautiful and useful result for us.

Select your Technical Indicator which is more comfortable to follow, analyse and good results on back test,moreover not complicated

No Indicators are sometimes even 70% profitable and no holy grail to make money.It is a combination of strategy,scrip,Time frame and the consistancy.

indicators are just secondary, even tertiary measurments of market movement itself and in my opinion detrimental to trade selection. Momentum indicators however have a place in assisting w/ letting a profit run. The majority of people wanting to learn to trade allow software to get in their way, they allow the computer to get between the market and themselves.

Nice advertisement, don't use titles like this without also indicating it's actually a product you're hawking. Pretty sure this violates the group's rules.

Each variation wasn't as small as 69 vs 70 on RSI but we did test things like buying at 40 and selling at 80 versus buying at 50 and selling at 80 and many iterations in between.

Indicators always follow price, never precede the next move.. If there's any indicator that could predict, it would be some formula that uses the main concepts of the technical analysis methods, talking about support & resistance at the first degree. The best approach to study this market is the lines you draw, the patterns you catch, and don't forget " your sense of the market/instrument " .

This sounds to me like a "unversity style" study. I mean when you trade for a living - own and/or customer money - the results count. Risk and reward after implementation. Different markets and assets, different time frames and compression, different methods, diversification, portfolio construction and risk mangement. Trying to survive, having no real bad year, creating steady, robust returns....

What is your aim by doing this? Education, selling studies, attraction customers on a platform?

Wondering: And for how much are you selling this pivotal information?

What about the out-of-sample behavior of these indicators? The in-sample overfit must be huge.

it is difficult to make a judgement on your claims without more information about how this study was constructed and how each indicator was tested. for eg. if a proportion of the testing was just trying different paramater values for the same indicator then this is clearly a curve fit. regardless anyone who seriously designs trading models knows that money is made from risk managmemt not indicators. you can use a coin toss, heads you buy and tails you sell which has no relationship to the market, will make money and have a positively skewed return distribution if you have a strong risk management strategy. so without more info it is hard to assess whether you work is meaningful or not.

I join Ron and agree with some other critic of the test. Actually , most of the people who commented criticized it.

I just wonder why this test attracts so much attention!

So you guys bought the Colby right?

Want to add two things:

(1) 17 different technical indicators is not that much. I could run said backtests on any backtesting platform like QuantConnect or Quantopian in minutes. I could download e.g. QuantConnect's LEAN opensource engine and run the complete set-up you describe in a matter of days.

(2) The extremely profitable results you describe are probably because of (i) overfitting or (ii) a massive amount of trades (any amount of equity grows insanely fast if the frequency of trades is high).

To give you an example, here is an indicator I backtested on QuantConnect that makes 1 million from a theoretical equity account of 10k in 10 years:

https://www.quantconnect.com/forum/discussion/967/daydreaming-from-10k-to-1-million-in-10-years#Item_5

As I explain in that post, the enormous gain is simply due to the high trading frequency and partial overfitting. It's not that impressive, really. Any slightly profitable short-term strategy will look impressive long-term!

Back tested models always look accurate and often sure things until you try to employ them in real time. Having spent 30 years specializing in option trading/analysis I would say if it looks to good to be true it is to good to be true. If it was in fact real returns rather than simulated via back testing models why would you tell anyone?

It seems to me that if tweaking a particular indicator gave you dramatically better results, then you are curve-fitting,or over-fitting, the data,creating over -optimization.This significantly increases the likelihood that the system will not work as well, if at all, in the future, because prices don't behave exactly the same as in the past. That type of indicator is exactly the opposite of what a system developer would look for in building a system because it indicates non-robustness! If you told me that you developed an indicator that gave astonishing results across a broad range of asset classes, time frames, and degrees of freedom, i would say you might have something there! Just my opinion.

Did you use walk forward optimization methodology?

Indicators work in a cyclic way,some times good results,sometimes bad because

markets have no continuity ,gap ups gap downs lead to loss.