Original paper
Abstract
This paper investigates the relationship of asset price determination via Google data and trading volume. We use weekly data from 2004 to 2010 for 30 international banks. Our study is the first which differentiate between Google’s search volume and Google’s search clicks. We find that asset prices are positively related to the growth rate of Google’s search, trading volume and the level of Google search clicks. Secondly, we find that the absolute level of Google’s search volume and Google’s search clicks behave differently regarding asset price dynamics. Google’s search volume, which measures long-run searches, is negatively related to asset prices and Google’s search click is positively related. We conclude that Google’s data contain important information for the identification of asset bubbles.
Keywords: Google Data, Asset Price Determination, Asset Bubbles
Trading rules
- Target assets: Stocks from the S&P 500 (excluding symbols prone to ambiguity such as ACE, COST, ZION).
- Data source: Google Insights Search.
- Update frequency: On a weekly basis, every Sunday.
- Focus: Number of searches for each stock’s ticker.
- Trading instructions each week:
- Rank stocks into five groups based on the prior week's search volume.
- Go long on high search intensity stocks.
- Go short on low search intensity stocks.
- Strategy for portfolio:
- Equally weighted allocation across stocks.
- Adjustments to be made every week.
Python code
Backtrader
import backtrader as bt
import pandas as pd
class GoogleSearchEffect(bt.Strategy):
def __init__(self):
self.search_intensity_data = pd.read_csv('search_intensity_data.csv') # Replace with actual data
def next(self):
current_date = self.data.datetime.date(0)
if current_date.weekday() != 0: # Only trade on the first trading day of the week
return
last_week_date = current_date - pd.DateOffset(weeks=1)
search_intensity_last_week = self.search_intensity_data.loc[self.search_intensity_data['date'] == last_week_date]
if search_intensity_last_week.empty:
return
search_intensity_last_week['quintile'] = pd.qcut(search_intensity_last_week['search_intensity'], 5, labels=False)
long_stocks = search_intensity_last_week[search_intensity_last_week['quintile'] == 4]['ticker'].tolist()
short_stocks = search_intensity_last_week[search_intensity_last_week['quintile'] == 0]['ticker'].tolist()
# Rebalance the portfolio
self.rebalance_portfolio(long_stocks, short_stocks)
def rebalance_portfolio(self, long_stocks, short_stocks):
current_positions = [pos.data._name for pos in self.getpositions()]
# Close positions no longer needed
for ticker in current_positions:
if ticker not in long_stocks + short_stocks:
self.close(data=self.getdatabyname(ticker))
# Enter long positions
for ticker in long_stocks:
if ticker not in current_positions:
self.order_target_percent(data=self.getdatabyname(ticker), target=1.0 / len(long_stocks))
# Enter short positions
for ticker in short_stocks:
if ticker not in current_positions:
self.order_target_percent(data=self.getdatabyname(ticker), target=-1.0 / len(short_stocks))
cerebro = bt.Cerebro()
cerebro.addstrategy(GoogleSearchEffect)
# Load S&P 500 stock data and add it to cerebro
# Exclude tickers with multiple meanings like ACE, COST, ZION
cerebro.run()Please note that you’ll need to provide the search_intensity_data.csv file containing the search intensity data for each stock in the S&P 500 index. The code assumes the file has columns ‘date’, ‘ticker’, and ‘search_intensity’. The stock data should also be added to the cerebro object.