This paper shows that dealer inventory plays an important role in liquidity transformation, especially for corporate bond ETFs. When these ETFs create shares in the primary market, corporate bond dealers deliver custom baskets of bonds containing a small subset of the ETF's underlying holdings. Using a novel dataset of corporate bond dealers and the delivered creation baskets, I show that ETFs primarily create with a small subset of dealers that hold larger inventories despite concerns of adverse selection and agency problems. I provide an explanation by exploring a channel where dealers with larger inventories have a relative advantage in delivering more illiquid bonds.  I propose a model to illustrate how dealer inventory can decrease the cost of liquidity transformation for ETFs, especially when information frictions are larger. Consistent with the model, dealers are more likely to overweight bonds in the delivered basket when they are in their inventory, especially when bonds are more illiquid. Additionally, looking at differences in the short-term bond returns, I find that inventory can decrease the cost of liquidity transformation by about 50%. This inventory benefit is strongest for more illiquid bonds and high-yield ETFs.

We examine active ETFs, particularly the recent innovation of less transparent active ETFs, to understand competition in the delegated asset market. Surprisingly, we find that ETFs cloned from mutual funds do not cannibalize the funds’ investor flows. The cloned mutual funds tend to have better reputations, giving the new ETF an advantage in attracting flows over their peers, even without better performance. We provide further evidence that investment companies introduce cloned ETFs for flow diversification. Our evidence suggests that some of the flows to the cloned active ETFs are driven by a difference in clientele from their mutual fund counterparts.

We craft a novel cybersecurity risk metric for firms by leveraging advanced sentiment analysis on corporate disclosures, testing our measure of cybersecurity risk not only using cybersecurity breaches but also a measure of the underlying cybersecurity vulnerabilities we construct by combining data on the software stacks used by firms and a government‐run database of software vulnerability scores. Our measure of sentiment analysis uses a unique methodology in grouping related passages and summarizing the relevant cybersecurity-related passages by extracting only key relevant and unique sentences. This helps reduce redundancy and noise in the sentiment measure. Overall, we find that a more extreme sentiment, regardless of whether it’s positive or negative, correlates with a higher underlying cybersecurity vulnerability as well as a higher likelihood of suffering from a data breach in the future. We compare this to older measures of cybersecurity risk and find they do not correlate at all to the real underlying cybersecurity vulnerabilities.

Bitcoin has become one of the most popular investment assets in recent years. The volatility of bitcoin prices in the financial market attracts both investors and researchers to study what predicts changes in its price. Existing works try to understand changes in bitcoin price by manually selecting features or factors. However, this trivial feature engineering consumes human resources without the guarantee that the assumptions or intuitions are correct. In this paper, we propose to predict changes in bitcoin price through understanding the patterns in the bitcoin blockchain transactions without feature engineering. We first propose a special type of graph, k-order transaction subgraphs, to capture the patterns. Then, we propose a framework we name Multi-Window Prediction Framework that utilizes machine learning models to learn the relationship between the patterns and the bitcoin prices. Extensive experimental results verify the effectiveness of these transaction patterns in understanding changes in bitcoin price and the superiority of the Multi-Window Prediction Framework to integrate multiple submodels trained separately on multiple history periods.