The current thesis presents three chapters in finance that investigate how a firm's characteristics affect its internal corporate financing decision as well as the external perception of the firm by the financial market. The paper advances empirical asset pricing by studying how the market's expectations about firms vary systematically with firms' profitabilities and the impact of these expectations on stock prices. Further, the paper adds to corporate finance by investigating new channels through which firms determine their capital structure and compensation contract for their executives.

Previous research indicated that sorting firms by their profitabilities generates sizeable cross-sectional variations in subsequent period stock returns. Firms that have high profitability this period tend to outperform low profitability firms in the next period. The first chapter of this paper investigates whether high profit firms have higher returns because they are fundamentally riskier in some way, or whether mispricing drives the return difference. The chapter presents evidence that the premium associated with profitability is difficult to reconcile with risk-based explanations but is consistent with expectation errors. Firms with a lower profitability prove more volatile, suffer greater drawdowns and are more sensitive to macroeconomic conditions. This means that the profitable firms are actually less risky by most measures and perform better during economic downturns. In addition, a monotonic relationship exists between profitability and forecast error. Stock analysts tend to be overoptimistic for low profitability firms relative to high profitability firms. Surprisingly, this mis-expectation can persist for as long as five years into the future. Furthermore, the profitability premium is mainly concentrated among firms about which the anlysts express the most optimism.

The second chapter, jointly authored with Francesco D'Acunto, Carolin Pflueger and Michael Weber, shows that the frequency with which firms adjust output prices is an important determinant of persistent capital structure. Using restricted BLS data, we constructed a measure of the individual firm's output price rigidity and matched it with financial information of the firm. Flexible-price firms choose higher financial leverage than inflexible-price firms, controlling for known determinants of capital structure. We rationalize this novel fact using a costly-state-verification model, where inflexible-price firms are more exposed to aggregate shocks, and hence face tighter financial constraints. The model predicts that bank lending, by providing monitoring, relaxes financial constraints and narrows the leverage gap between inflexible- and flexible-price firms. Consistent with model predictions, we show that inflexible-price firms increased leverage more than did flexible-price firms following the staggered implementation of the 1994 repeal of interstate bank branching regulations. Firms' frequency of price adjustment did not change around this deregulation, supporting a causal interpretation of price flexibility on corporate leverage.

In the third chapter of the dissertation, I investigate the role that innovation plays in executive compensation. Agency-based theory of optimal executive compensation literature has long puzzled over the empirical observation that executives are being compensated for industry performances that are outside the executive's control. Using patent-based metrics as the measure of innovativeness, I show that this “pay for luck” phenomenon is mainly concentrated among the most innovative firms and is stronger for firms in more innovative industries. I conjecture that motivating innovation might require a different contract design,and in some cases, “pay for luck” might actually be optimal in incentivizing experimentation.

Materials modeling is revolutionizing materials discovery paradigms through rationalizing the exploration of vast material design space. In general, materials modeling is built upon certain physics laws (e.g., computational simulations) and/or experimental data (e.g., machine learning). However, the state-of-the-art materials modeling is facing two grand challenges, i.e., (i) the high complexity of physics laws that govern materials properties, and (ii) the low informativity of experimental data. In order to address the two grand challenges of materials modeling, next-generation materials modeling aims to (i) make the physics simple to facilitate physics-driven modeling, and (ii) make the data informative to facilitate data-driven modeling.This thesis highlights the unparallel predictive power of integrating data-driven machine learning (ML) and physics-driven computational simulations to unlock a new era for materials discovery and for next-generation materials modeling: On the one hand, ML can assist in (i) developing empirical forcefields for accurate and computationally-efficient simulations, (ii) “separating the wheat from the chaff” in large amounts of complex simulation data to gain new insights or generate new knowledge of the underlying physics governing materials behaviors, and (iii) accelerating simulations by surrogate machine learning engines. On the other hand, simulation can generate large amounts of high-fidelity data that can be used to train machine learning models, which, in turn, can be validated by simulations. Both simulations and their integration pipeline with ML can be accelerated by leveraging automated differentiable programming engines and hardware accelerators. Overall, I envision that the “fusion” of simulations and ML models will unlock a new era in materials modeling—wherein traditional boundaries between physics and empirical models, knowledge and data, forward and inverse predictions, or experimental and simulation data would eventually fade. I hope that the present thesis will modestly contribute to stimulating new developments in that direction.

Post-earthquake structural damage assessment in reinforced concrete (RC) structures is an important concern within civil infrastructure research. Traditional structural damage detection (SDD) methods and advanced machine learning techniques often struggle to locate and quantify minor damages, especially in moderately damaged RC structures post-earthquake. This dissertation explores the potential of distributed fiber optic sensing (DFOS) technologies, including optical frequency domain reflectometry (OFDR), which offers high-resolution and microstrain-level accuracy, as an alternative for assessing internal structural damage. DFOS technologies provide a promising solution to the limitations of existing SDD methods, particularly for assessing minor damage in RC structures following moderate earthquakes.

The primary research objectives of this dissertation are: 1) to improve DFOS measurement capabilities by establishing a comprehensive methodology for assessing the sensitivity and durability of fiber-optic sensors and by developing improved algorithms for more accurate damage detection (e.g. crack width measurement and plastic hinge formation and progression), and 2) to create innovative damage indices that correlate residual and peak seismic deformations that enable DFOS application to quantify damage in post-earthquake scenarios. Furthermore, these methodologies have been applied and evaluated across various structural components like RC beams, beam-column joints, and moment frames.

The dissertation initially addresses the limitations of existing SDD methods by developing improved DFOS measurement techniques. Tests on six RC specimens evaluated the effectiveness of OFDR strain sensing in detecting concrete cracking and steel reinforcement deformation. Various fiber-optic cables, differing in structure, sensitivity, and survivability, were embedded within the concrete and reinforcing bars. A novel deconvolution method was developed, enabling accurate crack width measurements and revealing intricate bond-slip relationships, thereby establishing a foundation for advanced structural assessments.

Subsequent experiments focused on an innovative beam-column connection subjected to quasi-static cyclic loading. This test explored the potential of OFDR to capture detailed mechanical behaviors, such as beam curvature and slot opening and closing, enhancing our understanding of RC structures under cyclic loading conditions. Innovative damage indices were derived from the DFOS data, correlating well with the observed structural behaviors and demonstrating the applicability of DFOS in assessing RC structural damage.

Further explorations involve a testing campaign on RC bridge arch rib specimens subjected to varied axial and cyclic lateral loads, which provided new insights into the progression of cracking and plastic hinge formation. The development of a new damage index that correlates residual and peak seismic deformations showed the potential of DFOS for post-earthquake damage assessment, particularly when continuous monitoring is not feasible.

Finally, comprehensive testing on a two-bay, two-story RC moment frame tests the applicability of OFDR in a more realistic scenario. In this test, DFOS sensors provided accurate strain measurements across low to moderate drift ratio levels, aligning closely with strain gauges installed on reinforcing bars while offering detailed insights into microcracking at minimal drifts. By analyzing data from cables positioned on both sides of the neutral axis in beams and columns, the DFOS system effectively delineate the locations and extents of significant cracks and areas undergoing plastic deformation, withi findings corroborating visually observed damage states. Moreover, internal crack locations and widths, measured through the DFOS deconvolution algorithm, corresponded closely with DIC measurements, revealing a higher quantity of smaller internal cracks compared to fewer, larger surface cracks identified by DIC. The distributed strain measurements enabled a precise classification of the frame into uncracked, cracked and plastic regions, pinpointing critical areas where extensive plastic damage accumulates. The experiment underscore the potential of post-earthquake residual DFOS strain measurements to determine local damage levels sustained during peak seismic events, suggesting the need for further research to develop robust assessment tools for post-earthquake scenarios. This advanced understanding significantly enhances our ability to evaluate the integrity of RC structures post-earthquake. Overall, the findings from this research contribute new knowledge and new tools to improve the resilience of infrastructures against seismic threats, thereby supporting the goal of functional recovery in urban environments.

Development of drugs often fails due to toxicity and intolerable side effects. Recent advancements in the scientific community have rendered it possible to leverage machine learning techniques to predict individual side effects with domain knowledge features, such as drug classification. While several factors can be used to anticipate drug effects including their targets, pathways, and drug classes, it is unclear which domain knowledge is most predictive and whether certain domain knowledge is more important than others for different side effects. The goal of this project is to understand the predictive values of drug targets, drug classification (level 2 ATC codes), and protein-protein interaction networks (PathFX targets and network proteins) for the prediction of 30 frequently occurring side effects. We compared the prediction accuracy for individual side effects of trained models across five domain knowledge combinations and discovered that level 2 ATC codes have the highest predictive value across the domain knowledge features. Logistic regression coefficient analyses further suggest that side effects are significantly influenced by drug targets and drug classes, and not PathFX targets and network proteins. Our quantitative assessments may inform the development of safe and effective drugs by understanding the domain knowledge features underlying frequently occurring drug-induced side effects.

Biology frequently describes many abstract phenomena such as, but not limited to, DNA replication in the cell growth cycle, proton transportation in the energy production process, and ligand activation in the immune response system. Instructors often communicate such complex phenomena through symbolic representation. Students’ performance on exams and research papers limitedly gauge their understanding of the concepts. For example, it is uncommon for instructors to have time or opportunity to examine how well students understand some of the essential aspects of biology representations. Since the student population in the university system is becoming more diverse, we wanted to examine the relationship between English familiarity and interpreting one of the most common symbols used in biology textbooks, arrows. Through surveying 1969 students in multiple introductory biology courses, we evaluate the preciseness and consistency of students’ understanding of various types of arrows in the biological context. In our findings, English proficiency is correlated with students’ decisions when decoding the meaning of an arrow. In our exploratory research on the possible relationship between the visual representation and written description of arrows in biology, there exists enormous variation in understanding the meaning of arrows among all students, regardless of language status. We suggest the instructors in the biological field restate and enforce the specific and consistent usage of symbols.