Supertend - different values comparing with Tradingview

Question

Supertend - different values comparing with Tradingview

sivo1981 opened this issue 2 years ago · comments

BTCUSDT Perpetual - 1 day - ST (period = 10, multiplier = 3)

https://www.tradingview.com/chart/?symbol=BINANCE%3ABTCUSDTPERP

2022-07-01 H: 20926.4 L: 18948.8 C: 19272 Supertrend-Debut: undefined Supertend-TW: 23675.0
2022-07-02 H: 19456 L: 18959 C: 19240.9 Supertrend-Debut: undefined Supertend-TW: 23382.5
2022-07-03 H: 19645.4 L: 18780.1 C: 19305.9 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-04 H: 20350 L: 19046.3 C: 20226.6 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-05 H: 20749.2 L: 19291.6 C: 20170 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-06 H: 20667.6 L: 19753 C: 20554.1 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-07 H: 21830 L: 20244.9 C: 21613.4 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-08 H: 22579.9 L: 21175.2 C: 21587.9 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-09 H: 22000 L: 21309.6 C: 21580 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-10 H: 21594.2 L: 20600 C: 20853.8 Supertrend-Debut: undefined Supertend-TW: 23229.8
2022-07-11 H: 20853.9 L: 19870 C: 19954.3 Supertrend-Debut: 23570.9 Supertend-TW: 23229.8
2022-07-12 H: 20048.9 L: 19228.3 C: 19322.9 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-13 H: 20401.6 L: 18901 C: 20220.1 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-14 H: 20900 L: 19600 C: 20578.6 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-15 H: 21200 L: 20364.3 C: 20823.1 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-16 H: 21580 L: 20471 C: 21183.5 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-17 H: 21674.1 L: 20742 C: 20794.4 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-18 H: 22800 L: 20764.4 C: 22423.9 Supertrend-Debut: 22772.835 Supertend-TW: 23179.9
2022-07-19 H: 23817.9 L: 21575 C: 23385.9 Supertrend-Debut: 18871.8084752485 Supertend-TW: 18702.9
2022-07-20 H: 24277 L: 22912.4 C: 23221.5 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-21 H: 23441.1 L: 22340.2 C: 23144.5 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-22 H: 23750 L: 22500 C: 22675.1 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-23 H: 22990 L: 21933.5 C: 22440.2 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-24 H: 22999 L: 22256.4 C: 22570.6 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-25 H: 22656.8 L: 21232 C: 21300 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-26 H: 21334.4 L: 20689.6 C: 21241 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-27 H: 23100 L: 21031 C: 22941.1 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-28 H: 24194.6 L: 22583.1 C: 23836.2 Supertrend-Debut: 19743.14262772365 Supertend-TW: 19567.9
2022-07-29 H: 24500 L: 23426.3 C: 23771.1 Supertrend-Debut: 20168.013285324443 Supertend-TW: 20100.1
2022-07-30 H: 24700 L: 23505 C: 23629.9 Supertrend-Debut: 20328.376956791995 Supertend-TW: 20267.3
2022-07-31 H: 24187.3 L: 23222 C: 23290.1 Supertrend-Debut: 20328.376956791995 Supertend-TW: 20267.3

Supertrend values are quite different from what is published on Tradingview. Also direction is completely wrong based on this data.

Please advise if i'm doing something wrong beacuse i don't get the same values.

Dmitriy · Answer 1 · Fri Aug 05 2022 23:43:36 GMT+0800 (China Standard Time)

in tradingview default smoothing value for ATR indicator (used for sub calculations in SuperTrend) looks like this

smoothing = input.string(title="Smoothing", defval="RMA", options=["RMA", "SMA", "EMA", "WMA"])

i current indicators SDK default smooth value for ATR is 'WEMA'.

I published patch in debut/indicators@1.3.10, now you can change default smoothing type for SuperTrend indicator, please try to do like this:

const st = new SuperTrend(10, 3, 'RMA');

The third argument added in patch and allow configure smoothing MA type for ATR indicator, please check all of it and compare with TradingView, let me know about results, thanks for reporting

Dmitriy · Answer 2 · Sat Aug 06 2022 22:27:55 GMT+0800 (China Standard Time)

@sivo1981 can you check fixes?

Tadej Šivak · Answer 3 · Sat Aug 06 2022 22:37:30 GMT+0800 (China Standard Time)

I will check tomorrow ... out of office today.

Tadej Šivak · Answer 4 · Sun Aug 07 2022 04:12:21 GMT+0800 (China Standard Time)

I have tested this fix and the values are still not the same as on TW. I have used new RMA smoothing but looks like there is a problem when starting with OHLC from start of the month. I guess there is problem with historic data.

What do you think?

Tadej Šivak · Answer 5 · Sun Aug 07 2022 15:34:37 GMT+0800 (China Standard Time)

After further debuging i have found out that the problem is in ATR indicator with RMA smoothing.

Issue
ATR indicator uses getTrueRange function which doesn't work correctly.
TW description of getTrueRange:
handle_na (simple bool) How NaN values are handled. if true, and previous day's close is NaN then tr would be calculated as current day high-low. Otherwise (if false) tr would return NaN in such cases. Also note, that ta.atr uses ta.tr(true).

Currently getTrueRange returns undefined if prevClose is undefined. It should return high - low.

Issue
After manually patching 1. issue i have test ATR with RMA and the values are still not the same. Then i have tried ATR with SMA and values are OK. Looks like there is something also wrong with the RMA calculcation.

Dmitriy · Answer 6 · Sun Aug 07 2022 20:05:26 GMT+0800 (China Standard Time)

Thanks for clarify! I Add units for SuperTrend and True Range and bug was founded.

Check it with 'SMA' type smoothing.

@debut/indicators@1.3.11

Dmitriy · Answer 7 · Sun Aug 07 2022 20:12:47 GMT+0800 (China Standard Time)

Also about you comment

Currently getTrueRange returns undefined if prevClose is undefined. It should return high - low.

In any places what i see, i never see that calculation. When formula does not have prevClose should be undefined.

Tadej Šivak · Answer 8 · Mon Aug 08 2022 14:55:24 GMT+0800 (China Standard Time)

Tested Supertrend with SMA smoothing and it works OK. If i test with RMA it doesn't work correctly. This is beacuse the way RMA is calculated which is used in ATR. If prevValue in RMA is undefined then SMA for whole period should be used as first value but this is not the case here.

Can you check please.

Tadej Šivak · Answer 9 · Mon Aug 08 2022 15:28:01 GMT+0800 (China Standard Time)

I have attached excel comparison for ATR (period = 10) with SMA | RMA with Tradingview. See attached file.
ATR comparison.xlsx

Dmitriy · Answer 10 · Mon Aug 08 2022 15:57:36 GMT+0800 (China Standard Time)

RMA is not correct in TradingView, because its actually not RMA. Let's check it!

Debut RMA - it is "Relative Moving Average".
TradingView RMA (as we can see it from description)

it is Exponentially Weighted Moving Average . Correct name for TV RMA - is EWMA. Looks like they have wrong name for function. It is most popular problem with Moving Average. Nobody know how to do correct naming. For example RMA what is that means : Rounded Moving Average or Rolling Moving Average or Relative Moving Average - all this MA's have a different formulas to calc. This is big problem, but i do in my package with hard standards of naming and formulas validated many times.

Tadej Šivak · Answer 11 · Mon Aug 08 2022 16:01:31 GMT+0800 (China Standard Time)

The formula for RMA in TV is the same and correct. Source code:

//@version=5
indicator("ta.rma")
plot(ta.rma(close, 15))

//the same on pine
pine_rma(src, length) =>
	alpha = 1/length
	sum = 0.0
	sum := na(sum[1]) ? ta.sma(src, length) : alpha * src + (1 - alpha) * nz(sum[1])
plot(pine_rma(close, 15))

Dmitriy · Answer 12 · Mon Aug 08 2022 16:08:19 GMT+0800 (China Standard Time)

The formula for RMA in TV is the same and correct. Source code:

//@version=5
indicator("ta.rma")
plot(ta.rma(close, 15))

//the same on pine
pine_rma(src, length) =>
	alpha = 1/length
	sum = 0.0
	sum := na(sum[1]) ? ta.sma(src, length) : alpha * src + (1 - alpha) * nz(sum[1])
plot(pine_rma(close, 15))

Strange, description are different with implementation :) EWMA but.... RMA actually :)

Tadej Šivak · Answer 13 · Mon Aug 08 2022 16:19:09 GMT+0800 (China Standard Time)

RMA comparison (period = 10). See attached file.
RMA comparison.xlsx

Dmitriy · Answer 14 · Mon Aug 08 2022 16:41:39 GMT+0800 (China Standard Time)

RMA comparison (period = 10). See attached file. RMA comparison.xlsx

I have checked RMA, and nothing it there... formula are same. I have one idea. TradingView used pre-calculated values, that mean the calculations can have "machine precision error", when we have more price value then more error, can you try validate with some new tokens (less than 1 month history should they have) which price also not so big less than 1 - its ideal for testing. And send to me results please :)

Tadej Šivak · Answer 15 · Mon Aug 08 2022 17:16:14 GMT+0800 (China Standard Time)

1000SHIBUSDTPERP - Binance - 1 day

https://www.tradingview.com/chart/?symbol=BINANCE%3A1000SHIBUSDTPERP

RMA comparison (period = 10). See attached file.

1000SHIBUSDTPERP_RMA.xlsx

I think the problem is beacuse RMA function uses also SMA for first value. After N periods SMA is initialized but RMA start calculating after N periods. TV only uses SMA for whole period at start.

Dmitriy · Answer 16 · Mon Aug 08 2022 17:27:31 GMT+0800 (China Standard Time)

Can you validate RMA at long period i think the result are sync later.... may be. Another idea - TV calculate SMA for first value on history for ticker, all next values uses previous results from history start that affect the results and make it different. You need to check results from history begin.
check it here https://www.tradingview.com/chart/UqxDzKHm/?symbol=BINANCE%3AFLOWUSDT

Tadej Šivak · Answer 17 · Mon Aug 08 2022 17:33:42 GMT+0800 (China Standard Time)

You are correct. Calculating from start gives the correct result. I guess nothing we can do about it, if i start using indicator in between or at some specific date.

Thanks for your help.