The study of structural aspects of trade companies, especially those with e-commerce counterparts on Borsa Istanbul, is crucial for understanding market trends, investment opportunities, and economic stability. E-commerce's integration reveals how traditional commerce adapts to the digital economy, addressing the expanding digital consumer base and advancements in technology. Analyzing the interplay between these entities and their online counterparts sheds light on the challenges and opportunities of digital disruption. Multilayer networks offer a superior framework for analyzing intricate interconnections and interdependencies within a system compared to single-layer networks, revealing interactions and influences among different layers. This multifaceted viewpoint is essential for discerning behavior patterns and systemic weaknesses, fostering multidisciplinary collaboration. Applying this technique, policymakers can understand public sentiment, assess communication strategies, and enhance crisis management. This study focuses on companies in Borsa Istanbul's XTCRT index, constructing a 4-layer network using daily financial data and analyzing Twitter topics across six categories. The novelty lies in using multilayer financial network analysis to provide a broader perspective on financial relationships and correlations generated by listed companies, facilitating an understanding of interconnectedness and systemic risks.

The COVID-19 pandemic caused significant economic ramifications, disrupting global stock and financial markets. Borsa Istanbul, like many global markets, experienced sharp declines, particularly in sectors sensitive to changes in consumer behavior. Previous studies show that social media sentiment and discussions can significantly shift stock prices and trading volumes, sometimes predicting market movements. However, most studies focus on general market trends rather than specific sectors, and primarily use linear models, potentially overlooking the complex and nonlinear nature of online discussions. This study addresses this gap by integrating financial and social media data, offering a more nuanced understanding of investor behavior during the pandemic. The widespread use of social media platforms offers commercial companies opportunities to engage consumers, advertise products, and gain industry insights. Existing research highlights the significant influence of social media on investor behavior and the impact of news (regardless of accuracy) on market indicators. Previous studies often lack a sector-specific focus and advanced computational techniques, overlooking the multilayered impact of social media discussions on various aspects of financial performance. This study aims to address these gaps by utilizing a unique 4-layer network structure and advanced computational methods.

The study employs multilayer network analysis to explore interactions among companies within complex financial systems. This approach is vital for understanding intricate networks where companies are linked through various factors. This analysis helps identify underlying patterns and systemic risks, improving risk assessment and portfolio strategies. Using graph theory, a weighted graph is constructed where nodes represent assets and edges represent relationships. Methods like Minimum Spanning Trees (MST) and Planar Maximally Filtered Graphs (PMFG) are used to simplify the network structure. To analyze multilayer networks, a quadruplet M = (VM, EM, V, L) is defined, allowing a comprehensive view of financial relationships across different layers. The strength of a node (sa), weighted edge overlapped degree (oi), multilayer participation coefficient P(i), and the Z-score z(oi) are calculated. Finally, the entropy H of the multilayer network is calculated to measure its complexity. Topic classification, a crucial task in Natural Language Processing (NLP), is used. The study employs the Zemberek NLP library for processing Turkish language data, handling challenges like unprocessed data and agglutinative language characteristics. Deep learning, specifically LSTMs with GloVe word embedding, is used for topic classification. The Nonlinear Autoregressive Distributed Lag (NARDL) model is utilized to analyze financial data, capturing both short-term and long-term relationships and asymmetric effects. The NARDL model is represented mathematically, allowing for the distinct measurement of how positive and negative shocks affect the stock market differently. This model is particularly useful for exploring long-term cointegration and short-term dynamics. The dataset includes financial performance data for nine e-commerce companies listed in Borsa Istanbul's XTCRT index (BIMAS, BIZIM, CRFSA, DOAS, DOHOL, MEPET, MGROS, SOKM, TKNSA) and social media data from Turkish tweets related to COVID-19. Missing data gaps were filled using the autoregressive integrated moving average model. The study uses a 5-day sliding window for financial correlation networks and a 5-length sliding window for topic series analysis, employing the median instead of the mean due to skewness in the distributions.

The analysis of topological measurements of the 4-layer network reveals variations in mean multilayer participation coefficients (ranging from 0.7 to 1.0), indicating different degrees of interconnectedness among enterprises. Fluctuations may be associated with market stages influenced by the COVID-19 pandemic. Mean z-scores exhibit periodic variations around zero, with spikes signifying atypical activity. Multilayer entropies (ranging from 0.5 to 2.5) show a generally elevated degree of complexity, with decreases signifying periods of reduced complexity or predictability. The topic classification model, applied to a dataset of Turkish news articles and tweets, achieved high accuracy (over 90%). Graphical representations of average logarithmic returns of tweet volume for each topic (culture, economy, health, politics, technology, world affairs) reveal variations in discourse across different themes. NARDL analysis reveals significant short-term and long-term relationships between social media discussions and market behavior. There is a significant short-term inverse relationship between average multilayer participation coefficients and previous values. Economic discourse significantly influences market dynamics in both the short and long term. A strong negative correlation exists between multilayer network entropy and its past values. Positive political sentiment has a short-lived positive effect on market behavior. Technology-related discussions have a delayed significant impact on market behavior. Tables 1-12 present detailed NARDL results for different topics (culture, economy, health, politics, technology, world affairs), showing varying degrees of influence on both multilayer participation coefficients and network entropy. Generally, while past values of the network metrics significantly impact current values, the effect of social media discourse is more nuanced, often lacking statistical significance, except in the case of economic discourse.

The findings indicate a complex interplay between social media discourse and financial market performance in the Turkish e-commerce sector. The significant short-term inverse relationship between participation coefficients and previous values suggests dynamic market reactions. Economic discourse significantly impacts market dynamics, highlighting the sensitivity to public sentiment. The strong negative correlation between network entropy and past values indicates that market behavior is substantially influenced by historical patterns. Political discourse shows a more nuanced impact, with positive sentiment having a short-lived effect. The delayed influence of technology-related discussions highlights its increasing importance. The study's results highlight the significant responsiveness of e-commerce companies to economic and political social media content, leading to notable changes in market dynamics. These findings underscore the importance of considering public opinion and trust when designing public policies.

This study contributes to the understanding of how social media discourse influences the financial performance of e-commerce companies listed on Borsa Istanbul. The study’s findings highlight the significant role of public discourse, particularly in economic and political domains, on shaping market dynamics. Further research could explore causal links, use more granular data and sentiment analysis, and expand to other social media platforms.

The study's limitations include the relatively small sample size (nine companies), the focus on an eight-month period post-pandemic onset, and the use of data from only one social media platform (Twitter). These limitations affect the generalizability of the results, restricting the findings primarily to the e-commerce sector of Borsa Istanbul during the studied period.