SuperTrends on TradingView is not correct?

[l150039]

Hi team.
I find SuperTrends result is not correct.
I tested it by "bitcoin 1 day candle 2017-2020" and compared with tradingview indicator result but not matched. (bollingerbands result is correct! good!)

can you check? thanks!

[DaveSkender]

Can you provide more information?

1. Link to the TradingView indicator
2. How are you using the library? How much history, parameters, etc.

I’ll check it out. As a side note, TradingView often uses user-provided formulas, so it’s possible their formula is incorrect or that they made different assumptions.

Also, consider:

• this indicator has a warmup period where early values in the time series results will be less precise and should not be used; make sure you're using more than enough history
• the history you are using may not be identical to what is used in TradingView or another charting sites
• the first segment is always a 50/50 guess of direction and it should be discarded
• this library uses Close price as the trend breaker, some others may have customized it to use high/low "wicks" values

[DaveSkender]

On my initial review of the code, I'm not able to identify issues. Currently, I do not have enough information to determine if there's an error in this library, the TradingView indicator, or if the comparison is based on the same historical quotes.

If anyone can provide information to clearly show an error in this library's formula, please do. Otherwise, I'm going to assume that the TradingView implementation is errant or using "wicks" as the breakpoint, or that the provided historical quotes data is somehow different.

[l150039]

Hi Dave.

I test binance BTC-USDT (1day candle) Data. I will give you sample. (2017.8.17~2021.1.13 => 1246 days)
BTCUSDT_1d_1610582399999_1610550000000.json.txt

1. parsing data and try Indicator.GetSuperTrend.
   (lookbackPeriod_supertrends = 10, multiplier_supertrends = 3)

List<Quote> CandleHistory = new List<Quote>();

            StreamReader sr = new StreamReader("BTCUSDT_1d_1610582399999_1610550000000.json.txt");
            JArray CandleDataJArray = JArray.Parse(sr.ReadToEnd());
            sr.Close();
            for (int i = 0; i < CandleDataJArray.Count; i++)
            {
                Quote q = new Quote();
                q.Date = ConvertFromUnixTimestamp(CandleDataJArray[i]["openTime"].ToString());
                q.Open = Decimal.Parse(CandleDataJArray[i]["open"].ToString(), NumberStyles.Float);
                q.High = Decimal.Parse(CandleDataJArray[i]["high"].ToString(), NumberStyles.Float);
                q.Low = Decimal.Parse(CandleDataJArray[i]["low"].ToString(), NumberStyles.Float);
                q.Close = Decimal.Parse(CandleDataJArray[i]["close"].ToString(), NumberStyles.Float);
                q.Volume = Decimal.Parse(CandleDataJArray[i]["volume"].ToString(), NumberStyles.Float);
                CandleHistory.Add(q);
            }
CandleHistory = CandleHistory.OrderBy(x => x.Date).ToList()
List<SuperTrendResult> result_supertrends = Indicator.GetSuperTrend(CandleHistory , 10, 3).ToList();

2. and compare with binance tradingView (https://www.binance.com/en/trade/BTC_USDT)
   
   you can set 1day chart and supertrends by clicking where I colored yellow :)

Here is my result and Binance sample

(every value is not same OMG)
I tested "Indicator.GetSuperTrend" and "Indicator.GetBollingerBands" with binance BTCUSDT day value.
BollingerBands data is correct. it is same with binance tradingView's bollingerbands data, but supertrends is not (That is why I think my quote array is correct)

can you check it? If you need more information, comment please.
Thanks!

[DaveSkender]

Thanks for the extra info. I'll take a look in the next few days. I suspect this has to do with different approaches to the "initial guess" and warmup period at the beginning of the series. How well does the data compare at the end of the history series?

Generally speaking, as I mention in the documentation, the first segment before the first reversal is unreliable and should be discarded. I've been debating whether we should keep this data in the results or not as it can be confusing to users (see #306)

In this particular case, the formula relies on ATR values, which usually converge in precision after 100 periods. If you were to provide an additional 6 months of history, the date you selected above should have better matching. Charting sites will typically use more historical data [ATR data in this case] than shown, for this reason.

[DaveSkender]

At this point, I do believe the TradingView formula is wrong or is doing something unusual. Here's my mathematical proof:

The formula for the lower band is $SuperTrend = MidPrice - Multiplier \times ATR$ where $MidPrice = (High+Low)/2$. Using the actual values shown on the Binance TradingView chart on Dec 6th, as an example:

$$SuperTrend = (High+Low)/2 - Multiplier \times ATR$$ $$SuperTrend = (19420.91+18902.88)/2 - 3 \times 973.2082$$ $$SuperTrend = 16242.2704$$

The SuperTrend value on the Binance TradingView chart is showing an incorrect 16196.2300.

Using the data you provided, this library produces the expected value:

You can also review the manual calculations for this dataset in the Excel file and the source code in the repository.

[DaveSkender]

As a side note, I'm also looking at Barcharts.com and see quite a bit of difference between it and Binance/TradingView for SuperTrend. The Barcharts.com differences, I think, can be attributed to:

• different daily cryptocurrency prices, likely due to different daily cutoff times
• they also appear to exclude weekend data in their charts and calculations

I'm a bit skeptical with some of these newer charting sites, especially those that rely on unverified crowd-sourced or user-defined indicators. These variances are more likely a problem for less-common indicators like SuperTrend.

Generally speaking, given the 24-hour trading cycle for cryptocurrencies, I think it's also important to consider the source of data, how they're aggregating it into daily data, and whether or not they're using weekend data.

The good news is that with this library:

• you can control the source of quotes to meet your needs
• you can see the code used to calculate the indicators

[github-actions]

This Issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new Issue for related bugs.