None

none

Metacognition is defined as a person's awareness of the capabilities and vulnerabilities of their own cognition and also encompasses the actions that a person takes as a result of that awareness. The awareness and actions that a person takes are known as monitoring and control respectively. The relationship between accurate monitoring and improved control and performance has been borne out in multiple research studies. Unfortunately, people's metacognitive judgments are far from perfect; for low performers, that inaccuracy is most often in the form of overconfidence. Attempts to improve metacognitive monitoring and control have led to mixed results. The purpose of the experiments here was to examine whether participants could use confidence in their predictions to recalibrate subsequent performance predictions and to determine if improved metacognitive monitoring would confer benefits to metacognitive control. Would participants become less overconfident and would they then decide to study longer to improve performance? In three experiments, participants made predictions about their upcoming memory performance and reported their confidence that their predictions were accurate. Participants then adjusted their predictions so that they could be more confident the prediction was accurate. Experiment 1 served as a proof of concept ? it established that confidence judgments could be used to improve metacognitive monitoring accuracy. Experiment 2 explored the boundary conditions of the calibration improvement effect. The results revealed that continuous improvement in performance predictions was possible after reporting confidence. And finally, Experiment 3 showed that participants' improved monitoring accuracy did not influence metacognitive control, which in this study was allocation of study time. One possible reason why reporting confidence did not affect metacognitive control was that participants required feedback about the benefits of confidence judgments before the improved calibration effect would influence their decisions to allocate study time. Future research will examine the influence of reporting confidence and other interventions to improve calibration and performance.

This thesis examines important topics in corporate cash holdings and forecaster overconfidence. First, I provide an in-depth study of the interaction between intra-industry contagion risk and cash holdings. I develop a novel measure of a firm’s exposure to contagion risk that builds on the firm’s stock return comovement with other industry participants and separates contagion and competition effects caused by the expected financial distress of its industry peers. I show that high contagion-risk firms hold more cash because they face higher costs of external finance due to the potential decrease in their collateral values and the increased likelihood of their future financial distress caused by the net contagion effect. Second, in a co-authored paper with Drs. Jiaping Qiu and Chi Wan, we conduct a cross-country analysis to examine how financial development affects the reliance of corporate liquidity management on tangible capital. We find that financial development and better institutions lower the cash-tangibility sensitivity. This supports the view that financial development reduces the collateral role of tangible assets, thereby relaxing financial constraints of firms with low asset tangibility. This provides further firm-level evidence and sheds new light on the link between financial development and economic growth, as financial development promotes more efficient allocations of economic resources and hence facilitates investments and economic growth. Third, in a co-authored paper with Drs. Richard Deaves and Michael Schröder, we document using the ZEW panel of German stock market forecasters that weak forecasters tend to be overconfident in the sense that they provide extreme forecasts and their confidence intervals are less likely to contain eventual realizations. Moderate filters based on forecast accuracy over short rolling windows are somewhat successful in improving predictability. While poor performance can be due to various factors, a filter based on a prior tendency to provide extreme forecasts also improves predictability. / Thesis / Doctor of Philosophy (PhD)

No description available.

In this dissertation, I examine changes in risk-taking behavior, beliefs, and self-deception induced by changes in policy and behavior. Specifically, Chapter II examines player performance and risk-taking behavior in tournament environments which include eliminations in the middle of the tournament. I find that when players face elimination, they perform better and take risks more often. In addition, when facing elimination, players are more likely to have those risks pay off. Turning to the interaction between public policy and personal beliefs, Chapter III explores how public policy affects beliefs in the context of same-sex marriage. Exploiting the timing of the legalization of same-sex marriage, I find that legalization induces an increase in the proportion of people who have strong beliefs on same-sex marriage. I also find a substantial increase in measured state-level polarization due to legalization. Finally, Chapter IV presents the results of an experiment designed to uncover how self-confidence and self-deception change after performing dishonest behavior. In an online experimental laboratory, participants who cheated have higher confidence in their ability even when the opportunity to cheat is not present. In addition, participants who cheated, and were rewarded for cheating with a high reward, had higher beliefs in their ability. This dissertation includes unpublished co-authored material.

In the past, most researchers found investors tend to be overconfident when making investment decisions. This paper, under the assumption that investors have the disposition of overconfidence, tries to examine whether investors will display different degrees of overconfidence when facing different situations. Results have been found investors will have more confidence in processing the information the market has revealed to make investments when facing companies with high firm-specific risks. Therefore, firm-specific risks will influence overconfident investors¡¦ investment decisions. Furthermore, this paper tries to further divide companies under consideration into three groups according to their market share to discuss overconfident investors¡¦ behavior. Finally, this paper uses some quantitative variables to proxy public and private information to test whether investors in Taiwan tend to overreact to private information than to public information. The results have proven that investors¡¦ overreaction to private information in Taiwan comply with other behaviors of investors in other countries.

Classification is an important task in the field of machine learning and when classifiers are trained on images, a variety of problems can surface during inference. 1) Recent trends of using convolutional neural networks (CNNs) for various machine learning tasks has borne many successes and CNNs are surprisingly expressive in their learning ability due to a large number of parameters and numerous stacked layers in the CNNs. This increased model complexity also increases the risk of overfitting to the training data. Increasing the size of the training data using synthetic or artificial means (data augmentation) helps CNNs learn better by reducing the amount of over-fitting and producing a regularization effect to improve generalization of the learned model. 2) CNNs have proven to be very good classifiers and generally localize objects well; however, the loss functions typically used to train classification CNNs do not penalize inability to localize an object, nor do they take into account an object's relative size in the given image when producing confidence measures. 3) Convolutional neural networks always output in the space of the learnt classes with high confidence while predicting the class of a given image regardless of what the image consists of. For example an ImageNet-1K trained CNN can not say if the given image has no objects that it was trained on if it is provided with an image of a dinosaur (not an ImageNet category) or if the image has the main object cut out of it (context only). We approach these three different problems using bounding box information and learning to produce high entropy predictions on out of distribution classes. To address the first problem, we propose a novel regularization method called CopyPaste. The idea behind our approach is that images from the same class share similar context and can be 'mixed' together without affecting the labels. We use bounding box annotations that are available for a subset of ImageNet images. We consistently outperform the standard baseline and explore the idea of combining our approach with other recent regularization methods as well. We show consistent performance gains on PASCAL VOC07, MS-COCO and ImageNet datasets. For the second problem we employ objectness measures to learn meaningful CNN predictions. Objectness is a measure of likelihood of an object from any class being present in a given image. We present a novel approach to object localization that combines the ideas of objectness and label smoothing during training. Unlike previous methods, we compute a smoothing factor that is adaptive based on relative object size within an image. We present extensive results using ImageNet and OpenImages to demonstrate that CNNs trained using adaptive label smoothing are much less likely to be overconfident in their predictions, as compared to CNNs trained using hard targets. We train CNNs using objectness computed from bounding box annotations that are available for the ImageNet dataset and the OpenImages dataset. We perform extensive experiments with the aim of improving the ability of a classification CNN to learn better localizable features and show object detection performance improvements, calibration and classification performance on standard datasets. We also show qualitative results using class activation maps to illustrate the improvements. Lastly, we extend the second approach to train CNNs with images belonging to out of distribution and context using a uniform distribution of probability over the set of target classes for such images. This is a novel way to use uniform smooth labels as it allows the model to learn better confidence bounds. We sample 1000 classes (mutually exclusive to the 1000 classes in ImageNet-1K) from the larger ImageNet dataset comprising about 22K classes. We compare our approach with standard baselines and provide entropy and confidence plots for in distribution and out of distribution validation sets. / Doctor of Philosophy / Categorization is an important task in everyday life. Humans can perform the task of classifying objects effortlessly in pictures. Machines can also be trained to classify objects in images. With the tremendous growth in the area of artificial intelligence, machines have surpassed human performance for some tasks. However, there are plenty of challenges for artificial neural networks. Convolutional Neural Networks (CNNs) are a type of artificial neural networks. 1) Sometimes, CNNs simply memorize the samples provided during training and fail to work well with images that are slightly different from the training samples. 2) CNNs have proven to be very good classifiers and generally localize objects well; however, the objective functions typically used to train classification CNNs do not penalize inability to localize an object, nor do they take into account an object's relative size in the given image. 3) Convolutional neural networks always produce an output in the space of the learnt classes with high confidence while predicting the class of a given image regardless of what the image consists of. For example, an ImageNet-1K (a popular dataset) trained CNN can not say if the given image has no objects that it was trained on if it is provided with an image of a dinosaur (not an ImageNet category) or if the image has the main object cut out of it (images with background only). We approach these three different problems using object position information and learning to produce low confidence predictions on out of distribution classes. To address the first problem, we propose a novel regularization method called CopyPaste. The idea behind our approach is that images from the same class share similar context and can be 'mixed' together without affecting the labels. We use bounding box annotations that are available for a subset of ImageNet images. We consistently outperform the standard baseline and explore the idea of combining our approach with other recent regularization methods as well. We show consistent performance gains on PASCAL VOC07, MS-COCO and ImageNet datasets. For the second problem we employ objectness measures to learn meaningful CNN predictions. Objectness is a measure of likelihood of an object from any class being present in a given image. We present a novel approach to object localization that combines the ideas of objectness and label smoothing during training. Unlike previous methods, we compute a smoothing factor that is adaptive based on relative object size within an image. We present extensive results using ImageNet and OpenImages to demonstrate that CNNs trained using adaptive label smoothing are much less likely to be overconfident in their predictions, as compared to CNNs trained using hard targets. We train CNNs using objectness computed from bounding box annotations that are available for the ImageNet dataset and the OpenImages dataset. We perform extensive experiments with the aim of improving the ability of a classification CNN to learn better localizable features and show object detection performance improvements, calibration and classification performance on standard datasets. We also show qualitative results to illustrate the improvements. Lastly, we extend the second approach to train CNNs with images belonging to out of distribution and context using a uniform distribution of probability over the set of target classes for such images. This is a novel way to use uniform smooth labels as it allows the model to learn better confidence bounds. We sample 1000 classes (mutually exclusive to the 1000 classes in ImageNet-1K) from the larger ImageNet dataset comprising about 22K classes. We compare our approach with standard baselines on `in distribution' and `out of distribution' validation sets.

In the seven decades following the Investment Company Act of 1940 coming into force in the United States, the mutual fund industry has undergone dramatic changes including, some argue, a transition from stewardship to salesmanship with asset-gathering becoming the industry’s driving force. As fund managers incrementally assumed a more pronounced role in the investment fund industry, an emerging strand of finance literature focused on their characteristics and their potential impact on investment performance. While a large body of academic research concurs that fund managers cannot outperform systematically better than chance, there are also a significant number of studies that link the psychological characteristics of investors to their investment performance. Importantly, we know that fund managers, as a representative sample of professional investors, often have to operate under enormous anxiety and associated psychic pressures. In their effort to cope with these pressures and make sense of an immensely unpredictable and complex work environment, a wide range of psychic defences and behavioural biases may be triggered. The purpose of this research is to investigate, on the one hand, to what extent mutual fund managers are prone to overconfidence and associated behavioural biases such as self-serving attribution. On the other hand, the extent to which overconfidence, proxied by a wide range of variables including overoptimism, excessive certainty and excessive self-reference, may have any bearing on fund performance is of interest. The fundamental question is why, how, and through which mechanisms does overconfidence affect performance. The underlying research questions are motivated by three large areas of research: studies of mutual fund performance and persistence, studies of financial accounting narratives, and studies of professional investor psychology. I also explore how overconfidence is fundamentally generated and, in a sense, resorted to by fund managers as a defence mechanism against the psychic pressures of having to work in a highly intangible, complex and uncertain environment. Drawing on evidence from fund manager reports written for investors, I explain how they use the medium of narratives, and in particular stories, to make sense of what they do as fund managers and their added value for clients. I demonstrate how analysing fund manager commentaries, both through computer-assisted corpus-linguistic approaches and through the “close reading” method, sheds light on the link between fund manager psychology and investment performance. In particular, from the perspective of narrative analysis, I explain how fund managers write their reports in distinguishably different genres depending, among others, on their past performance record, fund size and investment style. In addition, I establish in a longitudinal study that the overall economic environment in which fund managers operate does influence the rhetoric of fund manager reports as well as the evidence for the Pollyanna hypothesis. My findings also suggest that excessive overconfidence is associated, to a large extent, with diminished future investment returns. While superior past returns are expected to increase fund manager confidence which, in turn, may introduce the overconfidence bias in the investment decision-making process and thus diminish returns (through inefficient stock selection, suboptimal market timing and other possible mechanisms), this is not a simple regression towards the mean. The asset pricing model employed in my empirical analysis, the Carhart four-factor model, controls for the effect of previous-year momentum, and my overconfidence measures are only slightly correlated with the momentum figures. Hence, one is led to the conclusion that the narrative-based variables used in this study indeed capture some aspect of the professional investor psychology, and are capable of enhancing the explanatory power of conventional asset-pricing models such as Carhart’s. In investigating the dynamic relationship between fund manager overconfidence and investment performance, the cross-sectional variations in my study demonstrate that superior past performance boosts overconfidence as measured by all proxies employed. In addition, there appears to be an inverted-U relationship between overconfidence and subsequent investment performance. In particular, a hedging strategy based on shorting funds with extremely overconfident managers and going long in funds with normally (over)confident managers, yields positive average returns. The impact of overconfidence on subsequent returns is robust across different investment styles, although it is stronger among growth-oriented funds. Incorporating average scores for fund manager overconfidence over longer periods yields similar results. In addition, fund manager duration appears to correlate with managerial overconfidence in the long term.

Data shows that many inventors continue to expend resources on their inventions even after they have received expert advice suggesting that they cease effort. Using a sample of inventors seeking outside advice from a Canadian evaluative agency, this paper examines how overconfidence, optimism, and illusion of control explain this fact. While overconfidence did not have a significant effect on inventor's decisions, illusion of control and optimism did have an effect. An additional interesting finding is that the more time people have spent working on inventions, the more likely they are to discount this expert advice.