Using Machine Learning to Understand the Dynamics Between the Stock Market and US Presidential Election Outcomes

Abstract

In this paper, we applied an explainable AI model (SHAP feature importance measures) to study the dynamic relationship between stock market returns and the US presidential election outcomes. More specifically, we wanted to study how the market would react the day after the election. AI models have been criticized as black-box models and lack the clarity needed for decision-making by different stakeholders. The explainable AI model we utilized in this model provides more clarity for the outcomes of the model. Using features commonly used by previous studies related to this topic, we find that the previous market direction leading up to the election and the incumbency information combined with the political affiliation are larger drivers for a 1-day post-election market return than sentiment and which party wins the election.

1. Introduction

“It is the economy, stupid”.

—James Carville

The relationship between US presidential elections and stock market performance is a topic that has been widely debated and analyzed by economists, political scientists, and investors. The US stock market has experienced notable fluctuations during election years. For instance, the period leading up to elections often sees increased volatility due to uncertainty about future policies and leadership. The stock market volatility index value typically goes up from 15% to 20% in non-election years to 20% to 25% during election years. Historically, the market tends to respond positively to incumbents who are expected to continue existing policies, while new candidates may provoke caution among investors, if the economy is doing well. On the other hand, there is a general perception that the stock market tends to perform better under a particular political party, with the conventional wisdom being that Republican administrations are better for the markets. However, the empirical evidence on this topic is mixed at best.

There are several theoretical channels through which US presidential elections could impact stock market performance. One prominent argument is that different political parties tend to favor contrasting fiscal and regulatory policies that can have varying effects on corporate profitability and investor sentiment. The Republican party is associated with more business-friendly, pro-market policies such as tax cuts, deregulation, and a hands-off approach to government intervention. In contrast, the Democratic party is often perceived as favoring greater regulation, higher taxes, and a more active role for the federal government in the economy. From this perspective, investors may anticipate better corporate earnings and stock returns under Republican administrations compared to Democratic ones.

Another potential linkage relates to policy uncertainty and market volatility. The election process itself can create uncertainty around the future policy direction, which may lead to increased market fluctuations as investors attempt to price in the potential implications. For example, campaign promises, and policy platforms of presidential candidates could signal major shifts in areas like trade, taxation, or regulation. This uncertainty may dampen investor confidence and contribute to stock market volatility in the lead-up to and aftermath of an election.

Furthermore, the election outcome itself may trigger policy changes that have direct impacts on corporate profitability and the broader macroeconomic environment. Shifts in fiscal, monetary, or regulatory policies under a new administration could affect factors such as consumer spending, business investment, inflation, and interest rates—all of which can influence stock market returns.

While the academic research on the connections between US presidential elections and stock market performance remains inconclusive, there are still important implications for investors and the broader economy to consider. For investors, the perceived relationship between elections and market returns may influence investment strategies and portfolio allocation decisions. Some investors may attempt to time the market or adjust their holdings based on anticipated electoral outcomes. However, the difficulty in accurately predicting market reactions to elections, as well as the potential for unexpected events, can make such strategies risky and prone to poor performance.

Given that deterministic methodology could not come up with a conclusive relationship presidential outcomes and stock market returns, we propose an alternative approach in this paper. More specifically, we developed a machine learning model that includes important features that include both party information and market data to predict the outcome of the presidential election. Once the prediction is made by the model, we then instruct the model to predict the reaction from the stock market the day after the election. The stock market is the most efficient market in the economy. Therefore, the price movements should incorporate all the wisdom of the crowd (Fama, 1970).

In this paper, we utilize the SHAP feature importance method by Lundburg and Lee (2017) to examine the various features that could affect the outcome of the US presidential election. SHAP feature importance allows us to gain insight into the following:

-

Model interpretability by showing the contribution of the various features into the model;

-

Visual representation;

-

A computationally tractable and partially understandable view into black box systems that shows the feature combinations that may be predictive.

By examining SHAP feature importance, we can assess the impact of various model features on predicting the market return the day following the election. We discovered that the most predictive features for post-election market performance were the market direction and whether the candidate had previously held office. While the market direction intuitively showed that a positive trend leading up to the election predicted a positive post-election market outcome, and vice versa, the incumbency of the candidate added an intriguing dimension. Specifically, if a Democratic candidate had prior incumbency, it predicted a downturn in the market the day after the election, whereas prior Republican incumbency predicted an uptick in the market.

The main contribution of this paper is the introduction of an explainable AI model to study the dynamic relationships between features and the predicted outcomes often missing in other machine learning/deep learning models. The remainder of the paper is organized as the following: Section 2 covers recent studies that examined the relationship between US presidential election and stock market performances. Section 3 discusses the methodology and feature selection. We then discuss the model’s outcome in Section 4 and conclude the paper in Section 5.

2. Literature Review

While the theoretical linkages between elections and markets seem plausible, the empirical evidence on this relationship has been mixed and inconclusive. Several academic studies have explored this topic, examining stock market returns during different presidential administrations and across party lines.

Hensel and Ziemba (1995) analyzes US investment returns during Democratic and Republican presidencies from 1928 to 1993. It finds significant differences in stock and bond returns based on political administration. Specifically, small-cap stocks experienced notably higher returns during Democratic administrations compared to Republican ones, largely due to fewer losses in the April to December period. In contrast, large-cap stock returns remained statistically similar across both party administrations. The study also highlights that returns on various asset classes were generally higher in the last two years of a presidential term, regardless of party. The authors tested several hypotheses regarding the performance of small-cap and large-cap stocks, as well as corporate and government bonds. They confirmed that small-cap stocks significantly outperformed large-caps during Democratic terms, while bond and cash returns were higher during Republican administrations. The research extends previous findings by emphasizing the small-cap advantage outside of January, and it notes that both small- and large-cap stocks tended to yield better results in the latter half of presidential terms.

Santa-Clara and Valkanov (2003) found that stock returns have been significantly higher under Democratic presidents compared to Republican presidents since 1927. The authors suggested that this outperformance may be attributable to more favorable macroeconomic policies under Democratic administrations, such as lower unemployment and higher productivity growth. The paper also suggests that government spending patterns and their impact on macroeconomic conditions are more important in driving cross-sectional stock returns. Specifically, the authors document that sectors with high government exposure, such as defense contractors and infrastructure-related industries, tend to outperform during periods of increased government spending, regardless of which political party is in power.

However, other studies have challenged this finding, arguing that the apparent Democratic “premium” in stock returns can be explained by other factors. Belo et al. (2013), using a comprehensive dataset spanning from 1930 to 2008, examined the relationship between government spending, political cycles, and the cross-section of stock returns in the United States and found that once you control for variables like the business cycle, monetary policy, and investor sentiment, the link between presidential party and stock returns disappears. The authors challenge the findings of prior studies that have suggested a connection between presidential administrations and stock market performance. The Belo et al. paper argues that the apparent Democratic “premium” in stock returns found in earlier research, such as the Santa-Clara and Valkanov (2003) study, can be explained by other factors beyond just the party affiliation of the president. The Belo et al. paper shows that after controlling for variables like the business cycle, monetary policy, and investor sentiment, the link between presidential party and stock returns disappears. They find no statistically significant difference in stock market performance between Democratic and Republican administrations. Furthermore, the authors find that the effect of government spending on stock returns is more pronounced around elections, as investors anticipate potential changes in fiscal policy. However, these effects are short-lived and do not persist over the longer term. Overall, the study by Belo and his colleagues challenges the notion of a systematic partisan influence on the stock market. The authors argue that broader macroeconomic and policy factors, rather than just political affiliation, are more crucial in determining cross-sectional stock market performance.

Snowberg et al. (2007) examines the impact of partisan politics on economic outcomes, using data from prediction markets and close elections. The authors argue that prediction markets can provide valuable insights into the market’s expectations about the economic effects of electoral outcomes. The study analyzes stock market returns in the days surrounding close presidential elections in the United States. The researchers find that the stock market tends to rise in the days immediately following a Republican victory, suggesting that investors anticipate more favorable economic policies under Republican presidents. Specifically, the authors estimate that a Republican victory leads to a 2% increase in stock prices over the three-day period surrounding the election.

The authors also find that the partisan impact on the stock market is larger for unexpected election outcomes. When the election result deviates from pre-election predictions, the market response is more pronounced, indicating that it is the surprise element of the outcome that drives the observed stock price movements. However, the paper notes that these short-term stock market gains following a Republican victory may not persist over the longer term. The authors caution against extrapolating the immediate market reaction into longer-term economic performance, as other factors, such as policy implementation and macroeconomic conditions, can ultimately shape market dynamics. Overall, the Snowberg et al. (2007) study provides evidence that partisan politics can have measurable effects on the stock market, at least in the short run.

One possible explanation for the mixed empirical findings is that the relationship between elections and markets is highly complex and context dependent. The specific economic conditions, policy platforms, and political dynamics at the time of an election can all play a role in shaping market reactions. Furthermore, the stock market may respond more to unexpected election outcomes rather than anticipated results.

Blau and Graham (2019) examines the stock market’s performance under Democratic versus Republican presidents in the post-2008 financial crisis period. Contrary to the mixed findings from earlier research, the authors find that the US stock market has performed better under Democratic presidents since the crisis. They attribute this to factors like increased government spending and more favorable economic policies implemented by Democratic administrations in the aftermath of the recession.

Chien et al. (2014) examines the relationship between the stock market’s reaction to presidential elections and the economic performance during the subsequent presidential term. Using data from 1900 to 2008 across 27 presidential administrations, the researchers test two hypotheses: (1) there is a relationship between GDP growth during a president’s term and the change in stock price immediately after the election, and (2) there is a relationship between unemployment rates during a president’s term and the change in stock price after the election. Their analysis also shows that the stock market’s reaction after an election has become progressively more accurate in predicting future GDP growth, but not future unemployment rates. The researchers found that Republican presidents tend to govern during periods where unemployment increases over their term, while Democratic presidents tend to see unemployment decrease over their term. Overall, the model appears to provide a good starting point for assessing the economic potential of new presidential administrations based on the market’s reaction to their election. Additionally, they find that the stock market has tended to respond negatively to Democratic presidential election wins, dropping in 10 out of 14 cases since 1900. This suggests investors may view Democratic presidents as less favorable for the economy compared to Republican presidents, at least in the short-term market reaction. The researchers conclude their model can help predict future economic performance based on the market’s assessment of the election outcome.

On cross-country comparison, Andrada et al. (2020) analyze the relationship between presidential elections and stock returns in Brazil. Their study finds that the election of left-wing presidents in Brazil is associated with lower stock market returns compared to right-wing presidents. The authors suggest this is due to investor perceptions that left-wing policies are less favorable for corporate profits and economic growth. Finally, Jia et al. (2021) examine stock market reactions to the two impeachment trials of former U.S. President Donald Trump. Their findings indicate that the stock market responded positively to events that increased the probability of Trump’s removal from office, implying that investors anticipated more market-friendly policies under a new administration. This highlights the importance of considering unexpected political events and their potential impact on investor sentiment and stock performance.

Hashim and Mosallamy (2020) explores the impact of presidential election outcome announcements on stock market return volatility in emerging markets, specifically Egypt, compared to developed markets like the United States. Utilizing a mixed-methods approach, the study incorporates both qualitative data and quantitative analysis, focusing on the EGX100 and S&P500 indices during significant elections. The authors aim to assess whether markets with different economic development levels exhibit similar efficiencies in responding to political events. The findings reveal that presidential elections do not significantly impact stock market volatility in either market. Although there are observable increases in abnormal returns and a decrease in volatility following election announcements, these changes are not statistically significant. This suggests that both markets efficiently absorb the news and integrate it into stock prices without significant shifts in volatility.

A key challenge in understanding the connections between US presidential elections and stock market performance is establishing clear causality. While there may be correlations observed between electoral outcomes and market returns, it can be difficult to determine the underlying causal mechanisms. For instance, it is possible that stock market performance could influence voter preferences and election outcomes, rather than the other way around. Strong economic conditions and rising equity prices may make voters more inclined to support the incumbent party, creating a feedback loop between markets and politics. As such, a deterministic model might leave out important features/connections. A parameter-free machine learning model might be a better choice when there are no clear-cut causalities.

3. Methodology

3.1. Feature Selection and Data Collection

Feature construction is critical to model building, this is especially true when there are low data volumes. The model is constructed using historical election data combined with market information. From the studies reviewed in the previous section, there are three main categories of features we decided to focus on: incumbency, investor sentiment, and market return. We look at these features once per election (every 4 years) with the various features looking back a different number of times. For incumbency data, we noted whether the currently running candidates had held office before, and we also looked at the last three election party winners. For market direction, we looked over three-to-eighteen-month windows leading up to the election, and for market return, we looked from the week before through the year before. Ideally, we would include as many features as possible and allow the model to show the features that are relevant. However, given the length of the data sample, adding too many feature risks resulting in model overfitting. We tried to strike a balance between having a useful number of features while avoiding the risk of overfitting. The primary features used for our model include the following:

• Incumbent party for both candidates and history of which party held office for the prior three elections;
• Market direction over 3–6-, 6–12-, and 12–18-month intervals prior to the election;
• Market returns over various time intervals in the year leading up to the election (between one week and one year leading up to election);
• Sentiment.

Once the data for the features are collected, we feed the data through a fully connected neural network with three hidden layers, which is employed to predict the outcome of elections (Democrat or Republican). Then, the party is added as a feature and used to predict the subsequent market direction (up or down). SHAP analysis is performed to measure feature importance, providing insights into the key drivers of the predictions. The features used in our model are summarized in

Table 1. Feature descriptions.

Feature	Descriptions
6–12_month_market_direction	1 if market was up from 12 months ago to 6 months prior to election else 0
prev_held_office_democratic	1 if the democratic candidate is an incumbent
3–6_month_market_direction	1 if market was up from 6 months ago to 3 months prior to election else 0
prev_held_office_republican	1 if the republican candidate is an incumbent
previous_party_3	1 if the 3rd most recent president was republican else 0
party	1 if the predicted party is republican else 0
day_before_365	Percent return from 365 days prior to election
12–18_month_market_direction	1 if market was up from 18 months ago to 12 months prior to election else 0
previous_party_2	1 if the 2nd most recent president was republican else 0
previous_party_1	1 if the most recent president was republican else 0
day_before_210	Percent return from 210 days prior to election
sentiment	A score of the favorability of the party in office
day_before_150	Percent return from 150 days prior to election
day_before_30	Percent return from 30 days prior to election
day_before_7	Percent return from 7 days prior to election

All data are collected via publicly available sources. We used LLM tools to analyze relevant news articles to determine the sentiment towards the party in office.

. A fully connected network is chosen to maximize the flexibility for data types as sufficiently large fully connected networks can approximate any continuous function. (Nielsen, 1987). The ease of implementation allows for simpler feature contribution and tends to offer better performance for small-scale tasks. Models are simply trained until losses are stabilized and are trained for a long time. Power et al. (2022) explains why we should train for a longer period. Future improvements can be made to see different initializations and see how the model performs. The model and training are less important than the feature selection as we want to showcase the features that tend to be important.

3.2. Model Architectures

We use a combination of neural networks and feature processing as the model architecture. The key components of the model are two neural networks, one fed into the other:

-

The first network predicts the winning party;

-

The second network, given the party prediction, predicts the market direction;

-

Both networks are fully connected with a downward cascade.

Figure 1 shows the model architecture graphically. The model’s implementation can be found in Appendix A.

We run two sequential neural networks: The left network is the party affiliation network, and the right network is the market outcome network. Each network consists of an input layer containing the 14 features we described for the left network and 15 features for the right network (the same 14 features plus the output party feature). The four “Gemm” (Generalized Matrix Multiplication) layers are the hidden layers, with Relu activation functions between each Gemm layer. The layers progressively reduce in size (e.g., 44 -> 22 -> 11 -> 1).

The Figure 2 presents the same model architecture in a DNN format, with three hidden layers between the features for predicting which party might win and how the victory might affect the market’s reaction.

4. Results and Feature Importance

The SHAP summary plot in Figure 3 illustrates the influence of each feature on the model’s predictions.

4.1. Model Output Analysis

The results from the SHAP analysis provide a visualization of the influence of each feature on the model’s predictions. The SHAP importance plot identifies the most influential feature to be how the stock market performs 6–12 months before the election, which shows a significant spread of SHAP values. The second most impactful feature is party incumbency, as it has the second highest SHAP values, suggesting that a candidate’s previous tenure as a Democrat plays a critical role in shaping the electoral outcomes. Additional important features include more recent market performance (3–6 months). This suggests that, while voters likely make their decisions based on the impact of current president’s policies on their investment portfolio more than 6 months out, they still care (to a lesser degree) about what happens to the stock market in more recent times. The burst of the 2001 DotCom Bubble and the 2008 Financial Crisis both happened towards the end of the election cycle, validating the importance of this feature. The other important feature is if the Republican party candidate previously held political office, which highlight differing influences of market conditions over shorter periods.

Features related to previous party affiliations and the current party show moderate importance, with varying effects based on their values. Time-based features, such as those reflecting returns from different days leading up to the election, generally exhibit less impact, with SHAP values clustering closer to zero. However, the feature “day_before_365” shows a wider spread, indicating potential relevance from events or conditions a year prior. The sentiment analysis feature appears to have a minor impact, with SHAP values concentrated near zero.

Overall, the findings reveal that market direction over different time frames and candidates’ incumbency status are the most crucial factors in the predictive model, while immediate temporal features and sentiment analysis play a lesser role in influencing electoral outcomes.

4.2. Discussion

The findings of this study have several important implications for future studies related to election predictions and stock market reactions. The strong influence of market direction over various time intervals suggests that economic indicators should be prioritized in election forecasting models. Analysts and political strategists may benefit from closely monitoring market performance, as it seems to correlate significantly with electoral outcomes. Additionally, the substantial impact of incumbency—specifically whether a candidate previously held office—highlights the necessity of incorporating political history into predictive models. This could lead to more accurate predictions by accounting for the advantages that incumbents typically enjoy.

Our study also demonstrates the importance of careful feature selection to avoid overfitting. Future models should aim to balance complexity with interpretability, focusing on a limited set of key features that have been shown to drive predictions effectively. While sentiment analysis was found to have a relatively minor impact, its incorporation could still add value, particularly in conjunction with other features. Future models might explore advanced sentiment analysis techniques to capture more nuanced public opinions, especially in volatile political climates.

Given that market conditions and political contexts can change rapidly, models should be designed to adapt dynamically. Incorporating real-time data feeds could help maintain accuracy as new information becomes available in the lead-up to elections. Furthermore, the findings should encourage further research into how these features perform across different election contexts, such as local versus national elections, and various demographic factors. Future studies could validate the model’s applicability in varying political landscapes.

5. Conclusions

In this paper, we proposed a predictive model for election outcomes and stock market reactions, using explainable AI techniques. The model integrates historical election data with market information, concentrating on three primary categories of features: party and incumbency, investor sentiment, and market return. While the ideal approach would involve including numerous features to identify their relevance, the use of excessive features could lead to model overfitting, given the small nature of the dataset. As a result, we used a limited number of features and showed how those certain features can be predictive. Specifically, we find that previous market direction leading up to the election and the incumbency information combined with the political affiliation are larger drivers for a 1-day post-election market return than sentiment and which party wins the election.