Investors and regulators are increasingly interested in identifying characteristics of entities that contribute to financial stability. A growing body of theoretical and empirical literature explores how interconnectedness affects the selling behaviour of banks during liquidation shocks (Allen, Babus, Carletti 2012; Acharya, Thakor 2016; Silva 2019). In times of stress, other non-banking institutions, such as insurance companies, also sell assets. These companies are subject to capital requirements based on risk and are linked to the broader financial system through their investments in specific asset types (Acharya et al., 2011). For a negative impact to occur, insurance companies don't need to fail in spreading risk throughout the system. It may suffice for them to "fire-sale" assets (Kartasheva 2014). Empirical research confirms that insurers’ trading behaviour during stress can impact prices and cause spillovers to other market players (Ellul, Jotikasthira, Lundblad 2011; Ellul, Jotikasthira, Lundblad, Wang 2015; Merrill, Nadauld, Stulz, Sherlund 2013; Manconi, Massa, Yasuda 2012).
Methodology
This paper addresses this issue by examining whether insurers with more similar portfolios are more likely to sell the assets they share. Using 2002-2014 data from the National Association of Insurance Commissioners, we calculate the cosine similarity of a pair of insurers' holdings. The cosine similarity is bounded between zero and one: a similarity of 1 indicates identical portfolios, while a similarity of 0 means completely different portfolios. We calculate the year-end similarities of each pair across broad asset classes as well as granular issuers. We demonstrate that portfolio similarity is related to insurer characteristics such as joint size, portfolio composition, and similarity in business lines.
We also show that our measure can accurately predict both the frequency and amount of sales made by companies with similar portfolios. We construct a measure for joint sales using information from insurer trades. This measure is the dot product of the vectors of quarterly net sales at the asset class level and the security issuer level. We find an inverse relationship between the portfolio similarity of a pair and their quarterly joint sales in the following year.
High-Risk vs. Low-Risk Assets
The overlap between insurers' portfolios may be attributed to liability matching needs, risk-seeking behaviours (Becker and Ivashina 2015), or both. We decompose each insurer's portfolio into high-risk and low-risk assets based on their likelihood to affect prices due to liquidity and credit quality. We calculate portfolio similarity across high-risk and low-risk assets and regress these similarities on liability similarity to determine the expected and unexpected portions of the similarities. We find that high-risk portfolio similarity supporting overlapping liabilities increases joint sales the most. Conversely, the "safe" or expected portfolio similarity across low-risk assets is negatively related to common sales. The effect of shared holdings on sales is largely due to increased risk-taking by insurers within the constraints of asset-liability management, making it more challenging for regulators to mitigate this effect.
Price Impact
We examine the change in the value of a pair's corporate bond holdings to determine if common selling by exposed insurers due to these shocks has a price effect. We calculate the average change in the yield spread of each pair's portfolio between the quarters before and after the shocks. We find that greater portfolio similarity increases yield spreads of pairs' joint corporate bond portfolios more for exposed pairs than for unexposed ones. Exposed insurers tend to sell more corporate debt and liquid assets like equity, mutual funds, and US government bonds. Thus, overlaps in insurers' holdings could lead to joint sales that may depress asset values under certain conditions.
We propose a portfolio-level similarity measure computed by comparing the average portfolio of an insurance company with other insurers from our sample. This measure helps identify institutions that may contribute to financial instability due to their divestment behaviours. It accurately predicts how much an insurer will sell in common with others, even after adjusting for size. Our measure is a valuable portfolio overlap tool for regulators monitoring potential systemic risk contributions from specific institutions.
Regulators are mainly concerned with corporate and sovereign bond markets where insurers play a significant role. Our findings could help regulators identify and monitor common investments made by insurers in these markets, where their divesting behaviour can amplify systemic risk.
This paper contributes to the growing literature on institutional investors' herding effects on asset allocation and liquidity. While prior studies have focused on corporate bonds, we propose a measure of commonality in portfolio holdings that encompasses an insurer's entire portfolio. This distinction is crucial as the sale of fixed-income securities (e.g., mortgage-backed securities) can spread risk among common holders (Merrill et al., 2013). Insurers can strategically trade between asset classes to mitigate the impact of price changes (Ellul et al., 2015). Our measure provides a comprehensive view of the relationship between portfolio similarity and expected sales.
Conclusion
We document a significant impact on the value of corporate bond portfolios due to insurers' herding effects, indicating a feedback loop between investors and asset prices, potentially destabilising the market. Our portfolio similarity measurement can identify institutions that may affect asset liquidation channels and systemic risk transmission.
Our measure of portfolio similarities can be calculated by any financial institution, regardless of whether they disclose their asset class or issuer holdings publicly or to regulators. This methodology can be used to analyse the assets of hedge funds, money market funds, and banks, allowing regulators to monitor potential common sale spillovers among various market participants. We find a positive correlation between portfolio similarity and SRISK, a measure available only to publicly traded companies. Therefore, our measure can be used alongside other risk metrics to monitor financial stability.
Related Content