Projects on Investors' Use of Information
- Information Sharing for Validation: Evidence from Social Networks (Job Market Paper, SSRN)
- Linguistic Complexity and Investor Horizons (with Brian Bushee, WP)
- Excess Volatility and Information Horizons (Solo-authored, WP)
1. Information Sharing For Validation: Evidence From Social Networks
|
Video Summary of the Paper
|
This study provides a novel perspective on why investors share information on social networks, challenging the conventional belief that they do so solely to influence public opinions and move market prices in their favor.
I propose that, in addition to the incentive to influence prices, investors share information to seek validation. Validation involves sharing information to better understand the true underlying state of the world and to reduce uncertainty in their own knowledge. My findings offer evidence that investors’ information-sharing behaviors on social networks align with both the incentive to influence and the incentive to validate, with the validation effect proving to be substantial. |
1.1 Why Do I Propose a New Incentive?
Traditional theories suggest that investors share information to influence market prices, which aligns with the one-way communication channels such as corporate disclosures.
However, social networks offer dynamic, two-way exchanges where investors can receive feedback and refine their beliefs. This interactive environment indicates that, beyond simply influencing market prices, investors may also be motivated by the need for validation.
However, social networks offer dynamic, two-way exchanges where investors can receive feedback and refine their beliefs. This interactive environment indicates that, beyond simply influencing market prices, investors may also be motivated by the need for validation.
I theorize this idea with an economic model (link). In this model, privately informed rational investors face a choice: share a noisy signal with rational but uninformed investors to collaboratively determine if the signal is informative about the asset's terminal value. This process captures the role of validation in investors' decision-making.
The model predicts that the informed investor will share and validate their noisy signal when there is sufficient uncertainty, represented by a low value of q in the model. The informed investors face a trade-off: uncertain but potentially larger profits from withholding the information, versus more certain, smaller profits from sharing it. They will choose to share if the benefits of validation are large enough to offset the loss of their informational advantage.
I theorize this idea with an economic model (link). In this model, privately informed rational investors face a choice: share a noisy signal with rational but uninformed investors to collaboratively determine if the signal is informative about the asset's terminal value. This process captures the role of validation in investors' decision-making.
The model predicts that the informed investor will share and validate their noisy signal when there is sufficient uncertainty, represented by a low value of q in the model. The informed investors face a trade-off: uncertain but potentially larger profits from withholding the information, versus more certain, smaller profits from sharing it. They will choose to share if the benefits of validation are large enough to offset the loss of their informational advantage.
1.2 How Do I Demonstrate the Existence of the Validation Incentive?
I investigate whether both incentives—to influence market prices and to validate information—are present within investor social networks. If these incentives exist, we would expect to see distinct patterns of information-sharing behavior when each incentive is expected to be stronger.
For example, the incentive to influence market prices is likely to be stronger when investors seek to buy or sell at more favorable prices. In contrast, the incentive to validate information becomes more pronounced during times of market uncertainty, as the expected utility from refining uncertain information increases.
For example, the incentive to influence market prices is likely to be stronger when investors seek to buy or sell at more favorable prices. In contrast, the incentive to validate information becomes more pronounced during times of market uncertainty, as the expected utility from refining uncertain information increases.
- H1: Investors with the incentive to influence are likely to share more actively on social networks when they want to liquidate their positions.
- H2: Investors with the incentive to validate are likely to share more actively on social networks during periods of high market uncertainty.
To test these hypotheses, I focus on the grain futures markets, which offer a good setting for identifying periods with strong incentives.
- The incentive to influence is likely higher around expiration dates and margin requirement increases, which trigger a surge in liquidation. (H1)
- The incentive to validate is stronger during exogenous shocks, such as the Russo-Ukrainian War or severe storms, when uncertainty is high. (H2)
Additionally, if the incentive to validate is significant, we should observe distinct effects in the information environments. That is, we would expect information to take longer to process and to improve in quality through validation.
- H3: Investors with the incentive to validate are likely to take more time to process new information.
- H4: Investors with the incentive to validate are likely to improve the quality of their information through sharing.
To test these hypotheses, I examine
- the speed of sentiment response following public report dates to assess how much time investors need to fully incorporate information. (H3)
- whether replies to the original validating posts refine information and have a stronger association with market prices. (H4)
1.2.1 How Do I Measure Incentives In Social Networks?
The setting allows me to identify periods with strong incentives, but to test the hypotheses, I also need to categorize information-sharing activities (e.g., posts on X.com) based on these underlying incentives.
To classify posts, I infer investors' motivations from their language use and apply factor and cluster analysis.
Specifically, I assume that investors aiming to influence market prices use more assertive and confident language (Allure, Allnone, Sure dictionary words), while those seeking validation tend to communicate in a more tentative and uncertain manner (Uncertain and If dictionary words).
To classify posts, I infer investors' motivations from their language use and apply factor and cluster analysis.
Specifically, I assume that investors aiming to influence market prices use more assertive and confident language (Allure, Allnone, Sure dictionary words), while those seeking validation tend to communicate in a more tentative and uncertain manner (Uncertain and If dictionary words).
1.2.2 How Do I Measure Information within Networks?
In the literature, sentiment has been shown to correlated with future performance, current returns, and the bias of qualitative disclosures, and I use aggregated sentiment in posts as a proxy for shared information. In particular,
As illustrated in Figure 3 below, posts on X.com exhibit varying levels of connectivity, and thus, differing degrees of influence over the network. The concept of centrality is key in identifying which nodes (posts) hold significant sway within a network, a principle widely used in social network analysis.
For this study, I utilize degree centrality, which is defined by the number of connections a node has.
- the sentiment of each post is derived using a RoBERTa-base LLM finetuned for twitter sentiment analysis tasks, and
- each post's sentiment is weighted by the importance of that post before aggregation.
As illustrated in Figure 3 below, posts on X.com exhibit varying levels of connectivity, and thus, differing degrees of influence over the network. The concept of centrality is key in identifying which nodes (posts) hold significant sway within a network, a principle widely used in social network analysis.
For this study, I utilize degree centrality, which is defined by the number of connections a node has.
In the X.com sample, the number of connections for each post is the sum of its likes, reposts, and replies. A post's weight in the network is determined by dividing its number of connections by the total connections of all posts in the network. To calculate the daily network sentiment, each post's weight is multiplied by its sentiment measure, and these products are summed to produce the aggregated sentiment for each day.
1.3 What Are the Main Findings?
Tables 5 through 8 in the paper reports the results from testing the hypotheses. In summary:
- Influencing posts tend to increase when investors are motivated to liquidate their positions, such as during expiration dates or margin requirement increases.
- Validating posts tend to increase during periods of exogenous shocks, like the Russo-Ukrainian war or severe weather events.
- Investors with the incentive to validate take more time to process and reflect on new information from public disclosures, while those aiming to influence respond more quickly, without such delays.
- Changes in the sentiment of replies to Validating posts most consistently explain price movements, compared to replies to Influencing posts or the original posts themselves.
Taken together, empirical analyses provide suggestive evidence that investors' information-sharing behaviors on social networks align with the two conjectured incentives and that the validation effect exists and is substantial.
2. Linguistic Complexity and Investor Horizons (with Brian Bushee)
Investors have varying resources and capacity constraints. Short-horizon investors, in particular, experience greater time pressures than their long-horizon counterparts. We hypothesize that these time constraints lead to informational disadvantages for short-horizon investors in two ways: they are deterred by high information acquisition costs and are more affected by substantial information integration costs, which result in the underutilization of available information. Using proxies for linguistic complexity, we find that short-horizon investors: (1) are less likely to invest in firms with opaque disclosures, (2) tend to ask fewer questions during conferences with high integration costs, (3) show less reactivity in their holdings to the information component of earnings calls, and (4) avoid investing in firms that are slow to disseminate information surrounding earnings calls, compared to long-horizon investors.
3. Excess Volatility and Information Horizons
This paper introduces differences in information horizons to study the excess volatility puzzle: Why do asset prices appear to be more volatile than is consistent with efficient market models? This paradox was first reported by LeRoy and Porter (1981) and Shiller (1980), and ever since researchers have made numerous attempts to resolve it. I propose a model that sequentially reveals information about firm fundamentals in the market and shows that, in the presence of varying information horizons, excess volatility can be explained. The model predicts that informed investors' demand decreases excess volatility and that investors with short-term information have a less negative impact on excess volatility because they are given a shorter time to act on their private information before it goes public. Next, using US market data (1987-2021), I find that excess volatility proxies decrease in institutional investors' demand and that these proxies are less negatively associated with short-term institutional investor trading activity compared to long-term investor trading activity. This study bridges the gap between traditional efficient market models and real-world data by taking into consideration that some pieces of information take a longer time to get impounded in market prices, which affects the type of investors who trade on the information.