Decoding the Hype
Finding Investment Signals in News Sentiment and Search Trends
Aria Ricci
May 30, 2025
Executive Summary
This report analyzes the predictive power of alternative data—specifically media sentiment and public search interest—on the stock returns of major technology companies from early 2020 to mid-2025. In an investment landscape where gaining an information edge is paramount, my analysis moves beyond traditional financial metrics to quantify the impact of public perception and hype on market performance.
My key findings indicate that while both news volume and search trends show a significant correlation with stock volatility, news tone (sentiment) is often a more potent and leading indicator of directional price movements. Positive media sentiment frequently precedes upward stock trends, whereas high search volume is more likely to coincide with periods of peak volatility rather than consistently predicting future returns. This analysis provides a framework for investors to use media sentiment as a supplementary tool for identifying potential market shifts and better understanding the narrative forces driving the tech hype cycle.
Introduction
Traditional stock analysis relies on a well-established universe of financial statements, economic indicators, and market fundamentals. While essential, these data points are often backward-looking and may not fully capture the real-time shifts in public mood and market narrative that increasingly drive volatility and asset prices, particularly within the fast-moving technology sector.
The rise of “alternative data” provides a new frontier for analysis. Digital news outlets and search engines generate terabytes of data daily that, when properly analyzed, offer a direct window into collective public interest and sentiment. This report seeks to answer a critical question for the modern investor: Can we find predictive, actionable signals within this digital “noise”?
I investigate the distinct roles of two key indicators:
Media Sentiment (Tone): The positive or negative sentiment of news coverage. Does the nature of the media conversation predict a stock’s direction?
Public Interest (Search Trends): The volume of online searches for a company or its products. Does the volume of public attention correlate with market moves?
By dissecting and comparing the influence of these two powerful data streams, this analysis aims to equip investors with a clearer understanding of how to interpret and leverage the hype cycle in their decision-making process.
Glossary:
- The GDELT Project
-
GDELT stands for Global Database of Events, Language, and Tone. It is a valuable and trusted open-source database containing statistics on broadcast, print, and web news from around the globe in over 100 languages. Check them out at https://www.gdeltproject.org/
- GDELT News Tone Score
-
GDELT uses an LLM to give a numerical tone score to each word in an article containing the keyword (ex. “NVDA”) and then uses the average of that score as the tone score for the entire document. Daily tone scores are an average of all document tone scores containing the keyword for a specified day.
- GDELT News Volume
-
Calculated as a percentage of total global online news coverage monitored by GDELT over a given day or hour.
- Google Trends
-
A source of data for search requests made to Google, operated by Google.
- Log Returns
-
A log value of daily returns, or log returns, is a way to measure the percentage change in an asset’s value over time using the natural logarithm of the ratio of its current price to its previous price. This method is often preferred in finance and statistics because log returns are additive over time, meaning the total log return for multiple periods is the sum of the individual log returns.
Methodology: Quantifying Influence
To ensure a robust and replicable analysis, I adopted a structured methodology encompassing data aggregation, metric definition, and statistical modeling.
Objective: To statistically measure and compare the influence of media sentiment and public search trends on the daily returns of selected technology stocks.
Scope & Timeframe: The analysis covers the period from January 1, 2020, to May 27, 2025, and focuses on a curated portfolio of technology stocks that represent a cross-section of the industry.
Data Sources:
Media Analysis: The GDELT Project was used to source daily data on news volume and average news tone (sentiment) for each company. GDELT monitors global news media in over 100 languages.
Public Interest: Search trend data was sourced from Google Trends, which provides a normalized index of public search interest for specific keywords related to each company. Data was downloaded using the gtrendsR then harmonized by the TrendEcon package to create consistent long-term time series.
Financial Data: Daily closing prices and log returns for each stock were retrieved from the Yahoo Finance API.
Analytical Process:
Data Aggregation & Cleaning: Daily time-series data was collected from all sources and aligned. Stock prices were converted to daily percentage returns to normalize the data.
Correlation Analysis: A Cross Correlation Analysis was performed to identify the strength and direction of the relationship between stock returns and my alternative data signals (news tone, news volume, search trends) over time.
Influence Modeling: To move beyond simple correlation and understand predictive influence, time-series models (such as Nonlinear Autoregressive Distributed Lag) were employed. This allowed me to test whether one time series is useful in forecasting another. This helps answer whether news and search trends lead or simply lag stock price movements.
Stocks in Analysis:
NVDA (Nvidia - semiconductor chips)
AVGO (Broadcom, semiconductor chips)
AAPL (Apple - cell phones, tablets, computers)
COIN (Coinbase - cryptocurrency exchange)
Caveats:
These findings are based on data from 2020-2025. Market dynamics can change.
The definition and measurement of “news tone” and “news volume” can influence results. The specific variables used here rely on GDELT’s method of capturing news tone and news volume.
This comparative analysis should serve as a guide for incorporating news-based indicators into prediction strategies, highlighting where news tone and search trends are a more potent signal.
Stock Performance
We’ve been largely living in a bull market for the past 5 years. The S&P 500 grew over 94% from 2020-2025, with an average annual return of 13.6%. This is higher than the historical average of 10%. The overall market’s strongest years were 2020, 2021, 2023, and 2024. In 2022, we experienced a bear market and new inflation concerns and tariffs in 2025 have clawed back some of the S&P 500’s earlier gains. As of June 1, 2025 the Dow, S&P 500, and Nasdaq are entirely flat for year-to-date gains, all down less than one percentage point of their values since January 1, 2025.
To put things in context, here are some key insights on the performance of each stock that will be analyzed in this project:
AAPL
Strong positive growth over the past 5 years.
A 5-year return of +151%.
49% market share of smartphone users in the US
AVGO
Significantly outperformed the S&P 500.
A 5 year return of +745.62%.
Strategically positioned as a supplier of AI infrastructure.
COIN
Growth has been negative over the past 5 years.
A 5 year return of -33.26%.
Caught up in the volatility of the cryptocurrency market and related trading laws.
NVDA
Remarkable growth.
A 5 year return of +1,488%.
Strategically positioned as a supplier of AI infrastructure.
Stock information obtained via Yahoo Finance
Google Trends vs Stock Price
A common issue associated with downloading Google Trends data in bulk is the “gtrends_with_backoff” error “429 - too many requests”. Some handy advice for anyone working with Google Trends data is to employ the gtrendsR “setHandleParameters” function to engage your VPN. I changed my IP address for every search and was able to gather data much more quickly than before.
Same-Day Search Keyword Correlation with Stock Price
Apple
Broadcom
Coinbase
Nvidia
The relationship becomes even more pronounced when each keyword is isolated against stock movement in the plots shown below.
Highest Correlated Keywords with Stock Price
Apple
Broadcom
Coinbase
NVidia
Keyword Correlation Analysis
There are two keywords that start to gain momentum ahead of their associated stock.
AVGO: “data center”, “semiconductor”
NVDA: “AI”
Coinbase has two keywords that seem to keep pace with the stock’s movement.
“bitcoin”
“crypto”
My analysis points toward the possibility that high-correlation keywords represent aspects of the businesses that are important to investors. Each company’s stock symbol or name ranks near the top of the correlation list but it’s surprising some aren’t more correlated to price.
Possible Reason: Uncaptured Data
A large percentage of a stock’s shares could be held by institutions and managed by people who don’t have to search the web for insights because they have their own proprietary info-gathering systems.
Fans of the stock could be engaging with news on other platforms. Coinbase is a great example. It has the lowest search volumes in my data set and a contributing factor could be that its main product is inherently decentralized and anti-establishment. Bitcoin watchers may check Google News for a portion of their intel, but they are also interacting on platforms that are harder to track like Reddit and TikTok.
Key Takeaways: Google Search Trends
Google Trends is a powerful tool for tracking public interest in a topic, however it is not a direct stock prediction tool. There is no guarantee that a spike in “data center” searches next week will lead to a rise in AVGO stock price. The strongest correlations appear with keywords related to the company’s core technology or products.
Google Trends offers a fascinating and potentially useful lens on the market by tracking public interest. Its best use for an investor is in identifying and monitoring the narratives and technological trends that are currently captivating market attention, rather than as a direct, mechanical “if this, then that” predictive tool.
Apple (AAPL): Mixed Signals
Insight: The signals for Apple are the most complex and demonstrate the limitations of this type of analysis. The strongest positive correlation is with “apple ai” (0.79), reflecting a more recent Apple product that Apple’s customers are tensely awaiting. Conversely, searches for its own ticker “AAPL” (-0.43), “apple stock” (-0.42), and flagship products like “facetime” (-0.40) have a negative correlation.
Visual Evidence: The “apple ai” plot shows a search trend that began to rise more recently with stock price, aligning with new broader public interest in AI tools. The “AAPL” plot shows that search interest for the ticker was highest during periods of volatility or decline (like 2020), hence the negative correlation.
Investor Takeaway: For a mature and well-known company like Apple, product search volume may not be a good indicator of stock performance. People search for “iphone” when they’re in the market for a new phone, and is not necessarily an investment signal even though iPhone sales bolster company profits. The strong negative correlation with “AAPL” suggests that retail investors may be searching for the ticker more frequently during times of bad news or uncertainty. This highlights the importance of ascribing context to keywords prior to tying them to stock analysis: why are people searching for a term?
Broadcom (AVGO): The Industry Demand Signal
Insight: AVGO’s stock price shows a very strong correlation with core business-to-business and industry-specific terms like “data center” (0.84), “semiconductor” (0.54), and “Broadcom” (0.82) itself. This differes from the more public-facing hype seen with NVDA.
Visual Evidence: The “data center” plot for AVGO is particularly telling. The steady accelerating rise in search volume for this term from 2022 onwards moves in tight concert with the stock price.
Investor Takeaway: Google Trends can be a proxy for unpacking underlying business demand and sector health. For a semiconductor company like Broadcom, rising search interest in the infrastructure it supports is a bullish signal. Investors can monitor trends for key industrial terms relevant to a company’s primary markets.
Coinbase (COIN): The Crypto Market Barometer
Insight: COIN’s stock price is (unsurprisingly) highly correlated with the general sentiment and interest in the cryptocurrency market. Terms like “crypto” (0.62), “coinbase” (0.54), and “bitcoin” (0.54) all show a moderately strong positive correlation.
Visual Evidence: The plot clearly shows that both the stock price and the search trends are highly volatile, with peaks and dips that often coincide. The significant spike in search interest in late 2021/ early 2022 corresponds to a peak in COIN’s stock price, and both declined together.
Investor Takeaway: COIN’s price is less about its specific corporate performance and more about the health of the entire crypto ecosystem. Google trends for major cryptocurrencies can serve as a real-time sentiment gauge for the sector. A spike in searches for “bitcoin” or “crypto” could signal increased trading activity, which is beneficial for Coinbase.
NVidia (NVDA): The AI Hype Indicator
Insight: NVDA shows an exceptionally strong positive correlation with search terms directly related to the Artificial Intelligence boom. Keywords like “AI assistant” (0.91), “AI” (0.88), “AI tools” (0.82), and even related technologies like “ChatGPT” (0.79) are highly correlated.
Visual Evidence: The explosive rise in search interest for “AI” and “AI assistant” beginning in late 2022 almost perfectly mirrors the stock’s rocket-like ascent.
Investor Takeaway: For a company like NVDA, which is at the epicenter of a major technological narrative, Google Trends acts as an excellent real-time gauge of AI hype and public sentiment. A surge in searches for the technology it powers can signal heightened retail and investor interest. The high correlation with its own ticker, “NVDA” (0.85), suggests that as interest in AI grew, so did direct interest in the company’s stock.
Search trends can be used to confirm a thesis or to understand the current public narrative around a stock, but it should never be the sole basis for a trade.
News Tone Interaction with Stock Price
GDELT’s tone scoring system ranges from -100 for extremely negative news to +100 for extremely positive news. All tone scores for news articles associated with “AAPL”, “AVGO”, “COIN”, and “NVDA” fell within the range of -6 to +6, staying far closer to neutral which is common for daily news.
Daily tone scores for each stock were placed into 5 bins:
Moderately Negative: -infinity to -1
Slightly Negative: -1 to -0.5
Neutral: -0.5 to 0.5
Slightly Positive: 0.5 to 1
Moderately Positive: 1 to infinity
As you’ll see below, the distribution of news tone scores was fairly even for all stocks, with AVGO and NVDA leading the pack in positive news, and AAPL and COIN leaning more neutral, overall.
Distribution of News Tone Scores
Apple
##
## Mod Neg Slight Neg Neutral Slight Pos Mod Pos
## 340 308 856 224 147Broadcom
##
## Mod Neg Slight Neg Neutral Slight Pos Mod Pos
## 54 92 520 507 702Coinbase
##
## Mod Neg Slight Neg Neutral Slight Pos Mod Pos
## 265 163 504 545 398NVidia
##
## Mod Neg Slight Neg Neutral Slight Pos Mod Pos
## 34 64 624 650 503NARDL Modeling:
Log Returns, News Tone, and News Volume
Dependent Variable: Daily Log Returns
Independent Variables: Daily News Tone, Daily News Volume
Decomposed Variable: News Tone was already decomposed into positive and negative values.
I used ADF testing on each variable to prove the data was stationary before modeling with NARDL. P-Values for News Tone, News Volume, and Log Returns were all below 0.01.
AAPL NARDL Findings
- News sentiment (positive and negative) has a statistically significant effect on AAPL stock returns in both the short and long run.
- News volume has a non-significant effect, implying it’s not the quantity but the tone of news that matters more.
- There is no evidence of asymmetry, meaning positive and negative tone have similar-sized impacts (no need to model them separately).
- A very high F-statistic (339) and R² (~51%) indicate a strong model fit, suggesting news sentiment can partially predict AAPL returns.
- Mean reversion in stock returns is evident—returns tend to reverse the previous day’s movement.
| Variable | Effect on Returns | Significant? | Interpretation |
|---|---|---|---|
| Positive Tone Score | Positive | Yes | Positive news sentiment is associated with higher AAPL returns both short and long-term. |
| Negative Tone Score | Positive | Yes | Surprisingly, even negative news sentiment shows a positive effect on returns, possibly due to overreaction or speculative trading. |
| News Volume | Positive | No | More news stories are not significantly linked to AAPL returns in this model. |
| Log Returns_1 | Negative (lagged) | Yes | Strong negative autocorrelation suggests mean reversion in stock returns (returns often move in the opposite direction of the previous day). |
AVGO NARDL Findings
Negative sentiment (tone) has a statistically significant and positive effect on AVGO returns—possibly reflecting contrarian investor behavior.
Positive sentiment and news volume (current and lagged) are not reliable predictors of AVGO stock performance in this model.
Lagged negative sentiment may slightly dampen returns, but the effect is weak.
A strong mean-reverting pattern is present, as shown by the highly significant and negative
log_returns_1.The model shows a strong fit with an adjusted R² of ~53% and an extremely high F-statistic (239).
No significant short- or long-run asymmetry was detected—meaning there’s no major difference in how positive vs. negative news affects returns in the long run.
| Variable | Effect on Returns | Significant? | Interpretation |
|---|---|---|---|
| Positive Tone Score | Positive | No | Positive news tone has no statistically significant effect on AVGO returns. |
| Negative Tone Score | Positive | Yes | Negative sentiment is significantly associated with higher returns, suggesting possible investor overreaction or contrarian trading. |
| Negative Tone Score_1 | Negative (lagged) | Marginal | Prior negative tone may have a delayed negative effect, but with weak significance. |
| News Volume | Positive | No | A higher volume of news is not significantly linked to return changes. |
| News Volume_1 | Positive (lagged) | No | News volume from the prior week shows no meaningful predictive power. |
| Log Returns_1 | Negative | Yes | Strong mean reversion in returns. Returns often move opposite to the previous week. |
The NARDL model for AVGO suggests that only negative news sentiment meaningfully influences stock returns positively. This may indicate a contrarian market response to bad news. Meanwhile, news volume and positive sentiment appear to have little predictive power for AVGO. The model also confirms mean-reverting price behavior, meaning returns often correct in the opposite direction of the previous week.
COIN NARDL Findings
News sentiment and volume have no predictive power in this model—neither positive nor negative tone, nor the amount of coverage, shows significant impact on COIN’s returns.
The only significant predictor is the previous week’s return, indicating a strong mean-reversion pattern.
The model fits fairly well with an adjusted R² of ~49.6% and an extremely significant F-statistic (250).
The cointegration test confirms a long-run relationship among the variables, but the drivers are not related to sentiment or news volume.
No evidence of asymmetry—positive and negative news appear to affect the stock symmetrically (i.e., neither direction matters).
| Variable | Effect on Returns | Significant? | Interpretation |
|---|---|---|---|
| Positive Tone Score | Positive (small) | No | Positive sentiment has no meaningful effect on COIN returns. |
| Negative Tone Score | Positive (small) | No | Negative sentiment also shows no significant relationship to returns. |
| News Volume | Positive (small) | No | News volume is not a useful predictor of COIN returns. |
| Log Returns_1 | Strongly Negative | Yes | Very strong mean reversion in returns. Price tends to reverse direction. |
For COIN, the NARDL model shows that news sentiment and volume do not significantly influence returns, suggesting that Coinbase stock may not respond predictably to news tone or media coverage. However, returns show a strong mean-reverting pattern, meaning past gains or losses are often followed by movement in the opposite direction. Investors may consider this behavioral dynamic rather than relying on news signals when evaluating short-term return potential.
NVDA NARDL Findings
Both positive and negative news sentiment significantly predict higher returns. This could reflect how NVDA is often in the spotlight—any attention may be good attention.
News volume doesn’t matter—it’s not how much news there is, but the tone that counts.
Strong mean reversion implies short-term price overreactions may correct in the following week.
The model is well-fitted (adjusted R² ~52.5%) and passes all diagnostics.
No asymmetry detected—positive and negative news have equal magnitude effects on returns.
| Variable | Effect on Returns | Significant? | Interpretation |
|---|---|---|---|
| Positive Tone Score | Positive | Yes | Positive sentiment significantly boosts NVDA returns, both short- and long-term |
| Negative Tone Score | Positive | Yes | Surprisingly, even negative sentiment is associated with positive returns. |
| News Volume | Neutral | No | News volume has no measurable effect on stock performance. |
| Log Returns_1 | Strongly Negative | Yes | Strong mean reversion in returns—NVDA often reverses the prior week’s move. |
For NVDA, both positive and negative news sentiment are strong predictors of higher returns, suggesting that the stock may respond positively to attention or controversy, regardless of the tone. While news volume itself doesn’t influence performance, returns tend to mean revert, highlighting opportunities to profit from temporary overreactions in price. This behavior positions NVDA as a sentiment-sensitive stock that reacts strongly to news tone rather than news quantity.
Key Takeaways: News Tone and Volume
1. News Sentiment Has Stock-Specific Effects
AAPL & AVGO: Only negative sentiment is significantly associated with higher returns (both in short and long run). This may reflect buy-the-dip behavior. Investors see overreactions to bad news as buying opportunities.
NVDA: Both positive and negative sentiment significantly predict higher returns. This suggests NVDA behaves like a “momentum darling,” where attention in any form fuels price moves.
COIN: Neither positive nor negative sentiment is statistically significant. This may indicate that COIN’s price movements are less sensitive to tone-based news sentiment, or that other factors like crypto trends dominate investor behavior.
2. News Volume Is Consistently Insignificant
Across all four stocks, news volume (both current and lagged) shows no significant impact on returns.
Implication: It’s not how much news circulates, but how it’s framed that drives short-term market response.
3. Strong Mean Reversion in Returns
All four models show highly significant negative coefficients on lagged returns (p < 0.001), suggesting a strong mean-reverting tendency.
Most pronounced in COIN and NVDA, highlighting potential for short-term reversal strategies.
4. No Evidence of Sentiment Asymmetry
Short- and long-run asymmetry tests were insignificant for all four stocks.
This means that positive and negative sentiment effects are symmetric—when they exist, they exert similar-sized impacts in opposite directions.
5. Prediction Strategy
AAPL: Focus on the direction and magnitude of changes in news tone.
AVGO: Be wary of initial reactions to negative news tone. The lagged effect is more aligned with typical expectations (negative news leading to lower returns). Positive news tone offers little predictive power.
COIN: Predictions for COIN should likely rely on other factors beyond the general news tone and volume measured here (e.g., cryptocurrency-specific news, Bitcoin price movements, regulatory developments, broader market sentiment towards digital assets).
NVDA: News tone is a critical factor. Monitor for significant changes in tone as leading indicators for NVDA’s price direction.
Investor reactions to news tone vary by stock: while AAPL and AVGO punish bad news, NVDA rallies on any kind of sentiment, and COIN appears unaffected. News volume itself doesn’t drive returns—but sentiment tone does, especially when stocks exhibit strong mean-reversion tendencies. These findings reinforce the value of tailoring sentiment-driven strategies to the behavior of individual equities.