Currency · Global · CFDs, forwards, futures, swaps · Monthly · Global Free

Investor Sentiment and Momentum Effect in Currencies

Academic paper

Investor Sentiment, Attention and Profitability of Currency Momentum Strategies

AuthorsPawee Maryniak

Institute

Wrocław University of Science and Technology
Wroclaw University of Economics and Business
?Uniwersytet Ekonomiczny we Wrocławiu
?Wroclaw University of Technology

Paper linkhttps://papers.ssrn.com/sol3/papers.cfm?abstract_id=2862161

Strategy in a nutshell

The strategy trades 71 currencies against USD, using past performance and the Baker-Wurgler sentiment index. Positions depend on sentiment: long Winners in low sentiment, short Losers in high sentiment, and both long/short in medium sentiment. The portfolio is equally weighted and rebalanced monthly.

Economic rationale

Investor sentiment drives currency momentum. Low sentiment pushes capital into safer assets, favoring long positions in foreign currencies, while high sentiment favors riskier assets, benefiting short positions. This behavior creates predictable patterns exploitable for returns.

Backtest performance

Annualised return5.8%

Volatility10.36%

Beta-0.02

Sharpe ratio0.56

Win rate37%

Full Python code

1from AlgorithmImports import *
2import numpy as np
3from dateutil.relativedelta import relativedelta
4#endregion
5class InvestorSentiment(QCAlgorithm):
6def Initialize(self):
7self.SetStartDate(2011, 1, 1)
8self.SetCash(100000)
9self.period:int = 21
10self.SetWarmUp(self.period, Resolution.Daily)
11
12self.symbols = [
13    "USDAUD", "USDCAD", "USDCHF", "USDCZK", "USDDKK", "USDEUR",
14    "USDGBP", "USDHKD", "USDHUF", "USDJPY", "USDMXN", "USDPLN", 
15    "USDNOK", "USDSAR", "USDSGD", "USDTHB", "USDTRY", "USDTWD", 
16    "USDZAR", "USDSEK"
17] 
18
19self.data:dict[Symbol, SymbolData] = {}
20self.sentiment_warmup_period:int = 12
21self.sentiment_history = RollingWindow[float](self.sentiment_warmup_period)
22self.max_missing_days:int = 31
23self.quantile:int = 6
24
25for symbol in self.symbols:
26    data = self.AddForex(symbol, Resolution.Daily, Market.Oanda)
27    data.SetLeverage(5)
28    
29    self.data[symbol] = SymbolData(symbol, self.period)
30
31# Import custom data 
32self.sentimet_index:Symbol = self.AddData(SentimentData, "sentiment", Resolution.Daily).Symbol
33self.recent_month:int = -1
34
35def OnData(self, data:Slice) -> None:
36perf:dict[str, float] = {}
37# store daily prices
38for symbol in self.symbols:
39    if symbol in data and data[symbol]:
40        self.data[symbol].Update(data[symbol].Value)
41    
42    if self.recent_month != self.Time.month and not self.IsWarmingUp:
43        if self.data[symbol].IsReady():
44            perf[symbol] = self.data[symbol].Return()
45if self.IsWarmingUp: return
46
47# monthly rebalance
48if self.recent_month == self.Time.month:
49    return
50self.recent_month = self.Time.month
51# check sentiment index data arrival
52if self.Securities[self.sentimet_index].GetLastData() and (self.Time.date() - self.Securities[self.sentimet_index].GetLastData().Time.date()).days > self.max_missing_days:
53    if self.Portfolio.Invested:
54        self.Liquidate()
55    self.sentiment_history.Reset()
56else:
57    sentiment_index:float = self.Securities[self.sentimet_index].Price
58    self.sentiment_history.Add(sentiment_index)
59
60    if not self.sentiment_history.IsReady: return
61    sorted_by_ret:List = sorted([x for x in self.data.items() if x[1].IsReady()], key=lambda x: x[1].Return(), reverse = True)
62    if len(sorted_by_ret) < self.quantile:
63        self.Liquidate()
64        return
65    quantile:int = int(len(self.symbols) / self.quantile)
66    winners:List[str] = [x[0] for x in sorted_by_ret[:quantile]]
67    losers:List[str] = [x[0] for x in sorted_by_ret[-quantile:]]
68    
69    percentile_33:float = np.percentile(list(self.sentiment_history), 0.33)
70    percentile_66:float = np.percentile(list(self.sentiment_history), 0.66)
71    long:List[str] = []
72    short:List[str] = []
73    # When the sentiment level in the previous month is low then only long position in winners is taken. When the sentiment level is high then only short position is taken. When sentiment level is medium then both short and long positions are taken.
74    if sentiment_index < percentile_33:
75        long = winners
76    if sentiment_index > percentile_66:
77        short = losers
78    else:
79        long = winners
80        short = losers
81    # liquidate
82    invested = [x.Key.Value for x in self.Portfolio if x.Value.Invested]
83    for symbol in invested:
84        if symbol not in long + short:
85            self.Liquidate(symbol)
86    # market execution
87    for symbol in long:
88        self.SetHoldings(symbol, 1 / len(long))
89    for symbol in short:
90        self.SetHoldings(symbol, -1 / len(short))
91class SymbolData:
92def __init__(self, symbol:str, lookback:int) -> None:
93self.Symbol:str = symbol
94self.History:RollingWindow = RollingWindow[float](lookback)
95self.Lookback = lookback
96def Update(self, value:float) -> None:
97self.History.Add(value)
98def IsReady(self) -> bool:
99return self.History.IsReady
100
101# Monthly return
102def Return(self) -> float:
103prices:List[float] = list(self.History)
104return (prices[0] / prices[self.Lookback - 1]) - 1
105
106class SentimentData(PythonData):
107def GetSource(self, config, date, isLiveMode):
108return SubscriptionDataSource("data.quantpedia.com/backtesting_data/index/baker_wurgler_sentiment_index.csv", SubscriptionTransportMedium.RemoteFile, FileFormat.Csv)
109def Reader(self, config, line, date, isLiveMode):
110index = SentimentData()
111index.Symbol = config.Symbol
112
113try:
114    data = line.split(';')
115    index.Time = datetime.strptime(data[0], "%Y%m") + relativedelta(months=1)
116    index.Value = data[1]
117except:
118    return None
119    
120return index