is ADX correct? (difference with TradingView)

[enriquepiatti]

I'm comparing the results from TradingView charts (with ADX built in indicator), same input period, same candles (this is ok because the SMA for example is returning same values in TradingView and with trading-signals), but the result for ADX is not the same!
Does somebody know why is that?
Do you know some other chart with ADX indicator to compare with a 3rd reference?

This is the code used by TradingView (in PineScript):

up = change(high)
down = -change(low)
plusDM = na(up) ? na : (up > down and up > 0 ? up : 0)
minusDM = na(down) ? na : (down > up and down > 0 ? down : 0)
truerange = rma(tr, len)
plus = fixnan(100 * rma(plusDM, len) / truerange)
minus = fixnan(100 * rma(minusDM, len) / truerange)
sum = plus + minus
adx = 100 * rma(abs(plus - minus) / (sum == 0 ? 1 : sum), adxlen)

plusDM and minusDM is the equivalent to
const { mdm, pdm } = this.directionalMovement(this.prevCandle, candle);
and they are returning the same values in TradingView and trading-signals., but the rest of the code is a bit confusing, can you note some difference in the way is calculated the ADX in TradingView? which one is the correct?

[enriquepiatti]

One difference seems to be the smooth function used, TradingView is using "rma":

Moving average used in RSI. It is the exponentially weighted moving average with alpha = 1 / length.

that gives some small difference in "plus" and "minus" from TradingView (corresponding to pdi and mdi from trading-signals).

But the important difference in values I think is in the last formula for adx, are TradingView and trading-signals doing different things?

[bennycode]

Hi @enriquepiatti. I think TradingView's ADX implementation is not standard compliant here. I have tested my ADX implementation and verified it with another implementation as you can see here: https://github.com/bennycode/trading-signals/blob/v2.0.1/src/ADX/ADX.test.ts#L36-L39

My implementation uses SMMA for smoothing. If it makes sense we can also add RMA (Relative Moving Average) as an option. Can you give me some input data and expected results for testing?

[enriquepiatti]

I think is not the only difference the SMMA vs RMA, here is the data for testing:

close:
[49320,48777,48708,47911.5,47883.5,47026,46860,46891.5,46897.5,46769,46956,46828.5,46928,47185,46832.5,46791.5,46870.5,46959,46999,46922.5,46981.5,47249.5,47179,46835.5,47261,47250,47213.5,46785.5,46917.5,47108,47162,47483,47185.5,47193,47410,47455.5,47340,47262.5,48123.5,48172,48190.5,48190,48333.5,48292.5,48911.5]

high:
[49370.5,49365,48825,48728.5,48018,48049,47244.5,47024,47072.5,47000,47040,47174.5,47400,47245.5,47399.5,47040,47050,47090,47078,47027,47046,47312,47422,47390,47300,47647,47490.5,47263,46974.5,47189,47249,47575,47582.5,47329,47495.5,47787,47539.5,47399,48275,48426.5,48273,48337,48370,48559,49213.5]
low:
[48844.5,48777,48644.5,47845.5,47600.5,46813,46735,46566.5,46685,46741,46702,46805.5,46272,46852.5,46828,46632,46505,46870,46718,46735.5,46834,46981,47116.5,46815.5,46335.5,47172,47203.5,46667,46573,46856,47027.5,46840,47125.5,46980.5,47182,47344.5,47230,46999.5,47178.5,48124,48004.5,48162,48140,48275,48185]

Expected result (this is the ADX value from TradingView for the last item): 24.51

RMA implementation in PineScript (according to https://stackoverflow.com/questions/48055618/how-tradingview-pine-script-rma-function-works-internally):

pine_rma(x, y) =>
	alpha = 1/y
    sum = 0.0
    sum := alpha * x + (1 - alpha) * nz(sum[1])

[bennycode]

Thank you for providing sample data. On Trading View there are multiple implementations of ADX from different authors. Which one are you using?

[enriquepiatti]

I'm using official ADX from TradingView, is marked as "built in", Average Directional Index.
It allows to see the pinescript code too, is the one I've posted (actually I have simplified it a bit to make it more clear what it's doing)

[enriquepiatti]

I couldn't replicate the same result from TradingView for RMA, not sure what I'm doing wrong, maybe you can see it.
This was my best approx:

class RMA extends MovingAverage_1.MovingAverage {
    constructor() {
        super(...arguments);
        this.prices = [];
        this.alpha = 1/this.interval; 
        this.rma = 0;
    }
    update(price) {

        this.prices.push(new big_js_1.default(price));
        if (this.prices.length > this.interval) {
            this.prices.shift();
        }

        if(this.rma === 0){
            this.rma = price;
        }
        // this.rma = this.alpha * price + (1 - this.alpha) * this.rma;
        this.rma = this.rma + (price - this.rma)*this.alpha;
        this.setResult(new big_js_1.default(this.rma));

[bennycode]

Can you reproduce the results from Tulip Indicators' ADX implementation? https://tulipindicators.org/adx

[enriquepiatti]

reproduce how? I don't know how to use Tulipan indicators, do they have an online version available?
Anyway, I think this is something related with TradingView, they must be calculating the ADX in a different way, probably is just that RMA instead of SMMA.
The problem I could't reproduce the values from RMA either!
and I think that's because RMA from TradingView is not using only the candles from the period/interval passed to the indicator, but is using all the candles, the interval is used for the calculation, but it requires more history to get the value used in TradingView (I will try that later).
But maybe could be a good addon to have an RMA and an ADX similar to TradingView too, is a very popular chart

[bennycode]

You don't have to run Tulip Indicators. They have their test input and test output listed here: https://tulipindicators.org/adx

I also have some other indicator implementations listed here: https://github.com/bennycode/trading-signals#alternatives

Usually, I check the test data of other libraries to know if my library is compatible or consistent with their output. That's an alternative approach to testing with TradingView.

[bennycode]

Hi @enriquepiatti, I have seen that you cross-posted your question: anandanand84/technicalindicators#244

I have done some research about ADX also and updated my ADX implementation so that you can use it with different moving average implementations for smoothing:

trading-signals/src/ADX/ADX.test.ts

Line 41 in 62ad238

const adx = new ADX(14, EMA);

Do you know to what RMA in TradingView is referring to? It could be "Relative Moving Average" but some people also think that it's a "Wilder's Moving Average": coin-unknown/Indicators#7

[Fc299684]

I had the same problem, currently what I do is get 200 OHLCV and convert to "ATRcandles", then create a new ATR(14) and implement forEach with the 200 candles in adx.update().
This brings me the same data that Binance shows for DMI +, DMI- and ADX.

  const ohlcv = await exchange.getOHLCV(setting.pair, setting.timeframe, setting.since, 200);
  const data = await pipes.createATRCandle(ohlcv);
  const adx = new ADX(14);
  data.forEach((candle) => {
    adx.update(candle);
  });
  const d = new Date();
  const values = {
    time: d.toLocaleString(),
    pair: setting.pair,
    timeframe: setting.timeframe,
    type: 'ADX 14',
    value: {
      pdi: adx.getResult().pdi.toFixed(6),
      mdi: adx.getResult().mdi.toFixed(6),
      adx: adx.getResult().adx.toFixed(6),
    },
  };
  console.log(values);
};

[bennycode]

It looks like the values are almost compatible with Binance's DMI indicator. However, the results are slightly different in the 5th and 6th decimal place. I wonder if this is a Binance rounding mistake or a Big.js rounding mistake. 🤔

[Fc299684]

It's true, I started testing and still not using websocket, and at the time of capture, I had not encoded the wait for the countdown to start analyzing the data! I'm still looking for a better performance

[bennycode]

@enriquepiatti today I noticed that the ADX return values of my lib are different from the calculations of Tulip Indicators (TI). I will update my implementation to match TI.

Here is the TI test data for reference:

• TulipCharts/tulipindicators@41e59fb
• https://github.com/TulipCharts/tulipindicators/blob/v0.8.0/tests/atoz.txt#L26-L29

[bennycode]

@enriquepiatti, @Fc299684 & @recidive: I have just released trading-signals v3.0.0 with a breaking change for ADX.

The getResult() method of ADX now just returns a single value which is the index itself (adx). If you want to get the +DI and -DI you will have to check the mdi and pdi members on the ADX instance. Make sure you only do this when isStable is true.

I also verified the implementation. By default ADX uses Wilder's Smoothed Moving Average to smoothen the DX values but you can change it if you like by providing another Moving Average class as second parameter to the ADX instantiation, e.g.:

import {ADX, EMA} from 'trading-signals';
const adx = new ADX(5, EMA);

My implementation is compatible with the Average Directional Movement Index from Tulip Indicators and returns the same results (given the same input). You can see this here: https://github.com/bennycode/trading-signals/blob/v3.0.0/src/ADX/ADX.test.ts

Best,
Benny

[Fc299684]

@bennycode Thanks, your awesome. I will test it!