Source-linked AI summary
Can Google Trends search queries contribute to risk diversification?
Ladislav Kristoufek
TL;DR
The paper examines whether Google Trends search queries can inform portfolio diversification. It weights stocks by search volume, discounting frequently searched stocks through a power-law rule. The ticker-based strategy reports a 163% out-of-sample cumulative profit versus 38% for the DJI index, while the paper also reports lower risk and outperformance against uniformly weighted portfolios.
Problem
The paper examines whether search-query information can support portfolio diversification without requiring information about correlations among DJI index components.
Source: Discussion · p. 4Method
The strategy uses Google Trends search volumes and a power-law weighting rule whose parameter controls the discount on frequently searched stocks.
Source: Results · pp. 2–3Results
163% out-of-sample cumulative profit for the ticker strategy compares with 38% for the DJI index; the reported search-based portfolios also reach lower risk and outperform uniformly weighted portfolios.
Source: Results · p. 4, Discussion · p. 5Takeaways & Limitations
Google Trends queries can be used in portfolio selection and risk diversification, with the ticker-based strategy outperforming the DJI index and uniformly weighted portfolio in the reported analysis.
Source: Discussion · pp. 4–5, Results · p. 4The stock/ticker strategy’s 88% out-of-sample cumulative profit versus 62% for the DJI index leaves a 26% difference over 4.5 years that the authors consider insufficient given potential weekly-rebalancing transaction costs.
Source: Discussion · p. 5
Abstract
from arXiv · showhide
Portfolio diversification and active risk management are essential parts of financial analysis which became even more crucial (and questioned) during and after the years of the Global Financial Crisis. We propose a novel approach to portfolio diversification using the information of searched items on Google Trends. The diversification is based on an idea that popularity of a stock measured by search queries is correlated with the stock riskiness. We penalize the popular stocks by assigning them lower portfolio weights and we bring forward the less popular, or peripheral, stocks to decrease the total riskiness of the portfolio. Our results indicate that such strategy dominates both the benchmark index and the uniformly weighted portfolio both in-sample and out-of-sample.
Results
The strategy weights DJI components using Google Trends search volume, with a power-law parameter controlling how strongly frequently searched stocks are discounted. The reported portfolios reduce risk and outperform uniformly weighted and DJI benchmarks, including out-of-sample.
Results
- The evaluation uses weekly returns for the DJI and its 30 components, comparing ticker queries with queries combining “stock” and a ticker symbol.The two query approaches cover different periods because of data availability, and some tickers are omitted for infrequent or ambiguous searches.Source: Results · pp. 2–3
- Portfolio weights use a power-law rule based on each stock’s search volume, with α controlling how strongly frequently searched stocks are discounted.Positive α lowers the weights of more frequently searched stocks; α = 0 gives uniform weights, while negative α favors more searched stocks.Source: Results · p. 3
- The study evaluates portfolio performance in-sample and out-of-sample using standard deviation and the Sharpe ratio.In-sample weights are rebalanced and returns realized in the same week; out-of-sample weights are rebalanced at week-end and returns realized the following week.Source: Results · p. 3
- For the ticker approach, standard deviation reaches its minimum at α = 0.1, while the Sharpe ratio peaks at α = 0.6 in-sample and α = 2 out-of-sample.The out-of-sample maximum Sharpe ratio is approximately 0.09, versus approximately 0.075 in-sample.Source: Results · p. 4
- Google Trends strategies achieve lower risk and outperform uniformly weighted portfolios, including out-of-sample; versus passive DJI buy-and-hold (2005–June 2013), the query-based strategy’s Sharpe ratio is 0.09 versus DJI’s less than 0.045.Source: Results · p. 4
- For the stock/ticker approach, the out-of-sample Sharpe ratio peaks at α = 0.6, while standard deviation falls through α = 1.3 before increasing slowly.The in-sample Sharpe ratio peaks at α = 0.2; the passive DJI strategy is dominated, though less strongly than in the other query approach.Source: Results · p. 4
Discussion
Google Trends portfolios dominate the uniformly weighted portfolio and DJI benchmark, though performance differs by query strategy. The ticker strategy’s out-of-sample cumulative profit is much stronger than the stock/ticker strategy’s, whose advantage may not cover transaction costs.
Discussion
- Ticker-based portfolios dominated the uniformly weighted portfolio and DJI benchmark, but the reported portfolios maximize Sharpe ratio rather than profit.The paper notes that profit-maximizing portfolios would use different weights.Source: Discussion · pp. 4–5
- The ticker strategy returned 163% cumulative profit out-of-sample versus the DJI index’s 38%, but neither strategy diversified away the Lehman Brothers loss.The in-sample ticker portfolio returned 109% cumulative profit.Source: Discussion · p. 5
- The stock/ticker strategy returned 88% cumulative profit out-of-sample versus 62% for the DJI index, a 26% difference over 4.5 years that the authors say may not cover transaction costs.Weekly portfolio rebalancing can bring substantial transaction costs; the ticker strategy’s 125% difference over 8.5 years was considered more promising.Source: Discussion · p. 5
- The study extends literature on search engines and financial-market profit opportunities by showing Google Trends queries can support portfolio diversification.The paper situates this contribution alongside prior work.Source: Discussion · p. 5
Data
Google Trends search data are weekly and updated daily for the running week, allowing Sunday rebalancing for the next week. The data’s normalization and changing index composition introduce specific implementation choices and comparison boundaries.
Data
- Google Trends reports weekly Sunday–Saturday series and updates the running week daily, so investors can access the information on Sunday to rebalance for the next week.Stock returns were reconstructed as weekly returns to match the query-series frequency.Source: Data · p. 5
- Because Google Trends normalizes series to a maximum of 100 and permits only five simultaneous search items, the study rescales groups of five, accepting possible rounding errors.Google Trends series are themselves rounded.Source: Data · p. 6
- The analysis uses DJI components as of 24 June 2013 without adjusting for five companies switched since 2005; the DJI comparison is illustrative.The authors focus on the potential of search queries for risk diversification rather than portfolio performance relative to the index.Source: Data · p. 6
Additional information
The figures compare portfolio-risk and Sharpe-ratio measures across search strategies and show portfolio-value paths against the DJI index. Their reported optimal discrimination parameters and relative performance vary by query type and evaluation setting.
Additional information
- For ticker-symbol searches, minimum variance occurs at α = 0.1 in- and out-of-sample, while the Sharpe-ratio maxima occur at α = 0.6 and α = 2, respectively.Standard deviations for varying α practically overlap across the two approaches.Source: Figure 1 · p. 10
- For ticker-plus-“stock” searches, minimum variance occurs at α = 1 in-sample and α = 1.3 out-of-sample, while Sharpe-ratio maxima occur at α = 0.2 and α = 0.6, respectively.Standard deviations for varying α are very close across the two approaches.Source: Figure 2 · p. 10
- The portfolio-value figure compares out-of-sample strategy paths with the DJI; ticker-symbol searches yield more than four times the DJI’s total profit, while ticker-plus-“stock” searches yield cumulative profit more than 0.4 times higher.The figure shows the ticker-only strategy on the left and the ticker-plus-“stock” strategy on the right.Source: Figure 3 · p. 11