The RSI is your king of oscillators. Mine and many others too. But I wonder if your methodology is useful to others. You show that, with the data your used, the (non-standard) parameters and your and your 5-day profit horizon, it was the best of the bunch. Does this mean much to anyone else?

When I was Head of Technical Analysis at Bloomberg many years ago, the RSI was the most popular purely technical indicator used by terminal owners. I did not count moving averages as they are used by many people who are not technical analysis or traders. The RSI was way out in front in usage. Given that it has been around for 40 years and that traders still love it and use it, it must be working for them. Traders are not stupid. They would drop it if it does not work. Popularity is a very good guide to what works. The RSI is king of the oscillators because, despite the advances in computing power, traders still love this old simple oscillator using their own data, parameters and timeframe.

Well said, Trevor Neil. True that my choice of parameters and profit horizon might not be useful to everyone, but hopefully the idea of how a group of indicators can be compared will be of some value.

I think the most important message here is that: "this data correctly uses the historical components of the respective indices". A feature rarely seen at any analysis software, and if seen often wrongly used.

Thanks Dion for your contribution today.

Do you have a favorite length of the indicator or chart increment you like?

Personally I like daily charts and trade based on very short term patterns. I don’t personally use these oscillators currently but was just curious about quantifying their effectiveness.

How well do technical indicators work with cryptocurrencies?

Thanks Dion. It would be interesting to know the results/payout if you had used the custom time period of 14 days (RSI,close,14) instead of 20 days (RSI,close,20). Would be greatful if you publish the results to test whether the default time period as intended by the creator of RSI still holds fort. Cheers

I noted that the rsi closely follows oscillations of the bullish consensus. Perhaps that fact explains its popularity.

Neppolian Pillai - I ran the analysis again with the RSI(14) included, and there results were: Dow30: 6.68, Nasdaq100: 2.33, S&P100: 4.17. So, still positive overall but far less effective than the 20 period. In my analysis I tried to give all of the indicators the same period (except ultimate oscillator which has 3 built-in periods) to be fair. Why would the 20 period outperform the 14? My thinking is that the 20 period will venture into oversold territory less frequently. So, you have fewer opportunities with 20, but those opportunities are of higher quality.

This kind of knowledge is more valuable than a gold.

So I see no reason why somebody, who know what indicator/technique is really works, will share this with the whole world.

Try to test relative volatility index (RVI), i think it is like rsi but better

Thanks for sharing !

Ivaylo Piskov I tested RVI but from what I read about it, it's intended more as a confirmation indicator. The results were: Dow 30: 3.39, Nasdaq 100: 2.05, and S&P100: -7.69

Yes, it is used more for confirmation instead of divergence, but for me it is more stable and i like it. In fact rvi is the only indicator i use

Ivaylo Piskov what platform are you using? I found some platforms calculation of RVI was blatantly incorrect, but interesting nonetheless :)

Range bound versus unbounded indicators are different. Comparing like indicators will yield more rigorous analysis. Different time parameters will yield different results for indicators, i.e. the Ultimate Oscillator issue. Third, the smoothing parameter and implementations yield differing lag and overshoot, i.e. StochD and StochK issues

@Dion..The reason that the 20 works "better" than the 14 is that the 20 has fewer inversions for the cycle that you are trading, hence fewer false positives.

Thanks Dion

@Dion Kurczek, i use TOS platform

This is a good threat. Asked earlier was, what perimeters do I use. I like consistency and robustness. I use, for example, 7 periods in the RSI with 65 and 35% lines. This shows me what I want to see from the RSI. The lower sample is because I want to expand the RSI vertically to catch more turning points and to remain 'overbought' or 'oversold' longer and therefore give me less false counter trend signals. It works for me in all timeframes and all securities. I therefore consider it (not optimised) but robust with these perimeters.

Awesome thread, thanks Dion Kurczek You also mentioned that you were going to add a video in YouTube for those who would like to test using your 30 day trial. Would you mind supplying those of us who are interested with a link?

Isaiah K Abraham Thank you very much! I have not yet made the video but as soon as it's completed I will post the link here.

Be aware though that all Oscillators are lagging and react only after price has moved.

I just wanted to introduce myself and my www.thinknum.com we sell alternative data points on 400,000 companies public and private companies with billions of data points updated daily . Let’s talk

1/2

Great work, Dion. The goal of the study is solid, though input-based variables will always greatly affect the results.

The main issue is that the 'default' overbought/oversold are still arbitrary values that were assigned by the old-school authors of TA decades ago. How do we know that RSI levels of 70/30 are more robust at identifying price reversals/regression than 80/20 or 90/10?

Making OB/OS levels more extreme or increasing RSI length will cut down on false positives, but will do at the cost of signal frequency and, ultimately, profitability. Finding an ideal balance for 'you' and whatever your desired trading parameters are (risk tolerance, liquidity factors, profit goals, leverage capabilities, strategy overlap) is always the goal.

2/2

We would be very interested to examine the same study if conducted using an algorithm that we developed to calculate "Dynamic Overbought/Oversold Levels" (DOBOS™). The optimal OB/OS levels are calculated by:

1. obtaining RSI values for the last "N" bars

2. separating the highest 10% values and the lowest 10% values; and,

3. calculating averages for the highest/lowest values.

This is a more objective approach using the indicator's own values to adapt to current price action ... rather than 'agreed upon' levels that are based upon tradition, not mathematical proof. Further, this allows for OB/OS levels to be calculated for unbounded indicators that do not have a set range (e.g. MACD, DMI).

Conducting this study with the full gamut of technical indicators (based on their own dynamic OB/OS levels) should yield some interesting, if not noteworthy, results!

Nice work Dion , I'd love to see more. I personally use more patterns and price action (oscillators and indicadores are a derivative of price) . I only use oscillators /indicator as confirmation , and I love them when they are in bullish/bearish divergences . My favourite always was RSI, then MACD and momentum, in that order. But for me 60% is price, 20% volume, 5% indicators/osc. 10% are other factors.

Have you ever tested heikin ashi as an indicator?

William Schamp Although the product I developed (Quantacula Studio) does include Heikin Ashi as a standard chart style, I must confess I have not tested it myself. What has your experience been, is there a specific test you'd like to see performed on it?

Zachary Gersch, CFTe I mocked up a 20 period RSI and dynamic OB/OS levels as you described in Quantacula Studio, and coded a quick backtest with an entry of RSI below DOS and an exit of RSI *crossing* below DOB. The results were pretty good. Originally I coded the exit to occur when RSI moved above the DOB, but changing it let the trade ride a little until RSI crossed *under* yielded higher profits without more volatility. I'll work on adding this concept to the analysis of the "edge" factor.

My study in 1987 concluded that the 12x3x3 Weekly Slow Stochastic gave the fewest false signals, which was my objective.

You only mention oversold? Did you also test overbought?

Would be interesting to see these for a market that doesn't have such a pronounced bias. Crude Futures or Currencies maybe?

Dion Kurczek - Sorry for the slow response, LinkedIn isn't sending notices to alert people of responses. I've just started finalizing the research I've been doing on them. I found them extraordinarily responsive and lag very little turning them into an indicator instead of the bars. I also use them on unconventional charting environments as well.

Richard Suttmeier - IMHO, Stochastics was one of the fist indicators I completely dismissed because it deadheaded. What I mean by that is it had a concrete ceiling and a concrete basement to the indicator. The best indicators are free flowing with absolutely no limit to their top or bottom structure. The last thing anyone needs is an indicator bouncing off a top ceiling in an extreme bullish move or bouncing off a bottom in an extreme bearish move.

'RSI - king of osc': I already examined all these and more awhile ago. You will get high stat. significance because of its definition it will be highly correlated. It has little forward-looking predictive power, which is you want. Using any variety of core stat ML techniques, you would get MACD, BB, and RSI or OBV as the most significant TA. Again, that is due to how they work. A prediction model based on those will not achieve meaningful returns in test. :)

FYI - I recently created an extension for Quantacula Studio that fully automates the calculation of six Indicator Edge Factors, read more about this here:

http://quantacula.blogspot.com/2018/02/indicator-edge-factors.html