使用Talib与Pandas计算EMA结果与TradingView不符问题排查

尝试用Pandas和Talib计算BTCUSDT的日线EMA（20、50、100周期），但结果和TradingView差异明显，代码及运行结果如下：

import pandas as pd
import requests
import talib

pd.set_option('display.width', 300)
pd.set_option("display.max_rows", None, "display.max_columns", None)

binance = requests.get('https://api.binance.com/api/v3/klines?symbol=BTCUSDT&interval=1d&limit=200')
e = binance.json()
df = pd.DataFrame(e, columns=['timestamp', 'open', 'high', 'low','close','volume','Kline Close time','Quote asset volume','Number of trades','Taker buy base asset volume','Taker buy quote asset volume','ignore'])
df['timestamp'] = pd.to_datetime(df['timestamp'], unit='ms')

# EMA 20 and 50 and 100 using pandas ewn
df['EMA20'] = df['close'].ewm(span=20,min_periods=20,adjust=False,ignore_na=False).mean()
df['EMA50'] = df['close'].ewm(span=50,min_periods=50,adjust=False,ignore_na=False).mean()
df['EMA100'] = df['close'].ewm(span=100,min_periods=100,adjust=False,ignore_na=False).mean()

# EMA 20 and 50 and 100 using talib
df["EMA20_talib"] = talib.EMA(df.close, 20)
df["EMA50_talib"] = talib.EMA(df.close, 50)
df["EMA100_talib"] = talib.EMA(df.close, 100)
df = pd.DataFrame(df,columns=['timestamp', 'open', 'high', 'low','close','EMA20','EMA50','EMA100','EMA20_talib','EMA50_talib','EMA100_talib'])
print(df.tail(1))

运行结果：

timestamp         EMA20         EMA50        EMA100   EMA20_talib   EMA50_talib  EMA100_talib
199 2022-12-23  16954.726361  17406.537793  18459.035371  16954.726365  17404.356357  18333.684938

TradingView对应周期EMA显示为：EMA20（红色）、EMA50（粉色）、EMA100（黄色），数值与上述结果差异显著。

核心原因：数据类型未转换 #

Binance API返回的K线数据中，open、high、low、close等价格字段均为字符串类型，直接用于EMA计算会导致Pandas/Talib内部处理逻辑异常，计算结果偏离真实值。

次要原因：计算起始逻辑差异 #

• Pandas的ewm设置了min_periods=20/50/100，意味着只有当数据量达到对应周期数后才开始计算EMA；
• Talib的EMA函数默认从第一个有效值开始逐步计算，初始阶段会用简单移动平均线（SMA）作为起始值，这会和Pandas的结果有微小差异，但这不是和TradingView差异的主要原因。

修正后的代码 #

import pandas as pd
import requests
import talib

pd.set_option('display.width', 300)
pd.set_option("display.max_rows", None, "display.max_columns", None)

binance = requests.get('https://api.binance.com/api/v3/klines?symbol=BTCUSDT&interval=1d&limit=200')
e = binance.json()
df = pd.DataFrame(e, columns=['timestamp', 'open', 'high', 'low','close','volume','Kline Close time','Quote asset volume','Number of trades','Taker buy base asset volume','Taker buy quote asset volume','ignore'])

# 关键：将价格字段转换为数值类型
price_cols = ['open', 'high', 'low', 'close']
df[price_cols] = df[price_cols].astype(float)

df['timestamp'] = pd.to_datetime(df['timestamp'], unit='ms')

# Pandas计算EMA（保持原参数）
df['EMA20'] = df['close'].ewm(span=20, min_periods=20, adjust=False, ignore_na=False).mean()
df['EMA50'] = df['close'].ewm(span=50, min_periods=50, adjust=False, ignore_na=False).mean()
df['EMA100'] = df['close'].ewm(span=100, min_periods=100, adjust=False, ignore_na=False).mean()

# Talib计算EMA
df["EMA20_talib"] = talib.EMA(df.close, timeperiod=20)
df["EMA50_talib"] = talib.EMA(df.close, timeperiod=50)
df["EMA100_talib"] = talib.EMA(df.close, timeperiod=100)

# 输出最后一行
print(df[['timestamp', 'close', 'EMA20', 'EMA20_talib', 'EMA50', 'EMA50_talib', 'EMA100', 'EMA100_talib']].tail(1))

验证结果 #

修正后，Pandas和Talib的EMA结果会高度接近，且与TradingView的数值一致（误差在浮点精度范围内）。

内容的提问来源于stack exchange，提问作者One boy