Public pension fund investments into hedge funds during the Great Recession: a network analysis

State public employee retirement systems in the United States control over $5 trillion in assets, with large investments in alternative asset classes like hedge funds. This paper studies public pension fund investments into hedge funds in the period immediately surrounding the 2008 Financial Crisis and the Great Recession. Using micro-level investment data, the paper examines the characteristics of this network of investment flows and compares these networks to similar investments made by private pension funds. We find that there is considerable overlap among public pension funds in their choice of hedge fund investments. While the analysis is limited by data availability, the paper provides a novel study of actual investment decisions rather than aggregate portfolio outcomes.


Introduction
In the United States, state public employee retirement systems have over $5 trillion in assets in their pension funds that are meant to provide retirement benefits for over 25 million individuals employed by state and local governments (NASRA 2023). About a quarter of these assets are typically invested in alternative investments by state pension funds, and this percentage has rapidly increased in the past 15 years (Banta 2022;Cliffwater 2022). While the shift in public pension portfolios from traditional fixed income and equity investments to alternatives has been fairly widespread, it is unclear whether the strategy ultimately enhances investment performance and stability. It is difficult to determine relative performance due to differences in reporting norms across state systems, specifically in the way that portfolio returns are reported as either gross or net of investment fees. The performance of these investments is critical for budget planning and viability for state and local legislatures.
One of the main challenges in analyzing public pension fund investment performance is the diversity of investment strategies and available alternative investment options. This article focuses on examining public pension fund investment strategies in hedge funds, one of the most significant and contentious alternative investment avenues, during the critical period of the Great Recession. The article uses investment-level data rather than portfolio-level data to determine the degree of homogeneity between public pension funds in their hedge fund investments and the ability to diversify across hedge funds. The study compares the network of public pension investments in hedge funds to that of private pension funds. Although the data are incomplete, the results suggest that there are similar hedge fund investment patterns across the included public pension funds, even as public pension funds invest in a greater number of hedge funds than the comparison set of private pension funds.
This study offers a critical contribution to public investment research. First, it provides a unique look at micro-level data on state retirement fund investments, rather than aggregating overall alternative investments. Second, it demonstrates the extent to which disparate state systems have overlapping investments in specific hedge fund families, thereby shifting the question of pension fund performance from a statelevel budgeting issue to a systemic national issue. Finally, the study suggests additional questions surrounding investments into hedge funds, including the extent to which external managers with tenuous accountability to state taxpayers exert control over multiple system funds.

Public pension funds in the United States
There are over 2,000 pension systems at the state and local level in the U.S. (GAO 1996). Public retirement systems have transformed from a form of social welfare to retired workers to a vital incentive for government employees. The majority of public pension funds operate as defined benefit systems (though increasingly there are hybrid and defined contribution plans), where the public employer and sometimes public employees make regular contributions to the fund. The assets are then invested and used to pay benefits to current retirees, with the government sponsor obliged to fulfill promised benefits. Retirement benefits are calculated using a schedule of factors, such as tenure, salary and retirement age, and are usually updated for cost-of-living adjustments through formulas tied to inflation.
In addition to retired public employees, other stakeholders include the pension sponsor and administrators overseeing the fund. If the funds face insolvency and cannot pay monthly benefits, jurisdictions may issue government bonds, raise taxes or shift other government revenue to the retirement systems and as such, taxpayers are ultimate liable. Furthermore, taxpayers have a stake in maintaining and attracting quality government workers, as retirement and health benefits are often a significant attraction for state service. Independent agencies, managed by a board of trustees, often administer public employee retirement systems. The jurisdiction's legislative body may retain control over the budget, with the delegation of authority these agencies varying in degree. The composition of the fund boards varies, with trustees chosen by appointment, elections by the general populations or elections by current and future retirees. The level of fund beneficiary capture thus differs by fund. Investment, actuarial activities, benefit levels and contribution payments/budgets are key areas of decisionmaking, and few funds altered the delegation of benefits and budget control in any measurable way in the decade before or after the time-period studied here.
Much of the existing empirical pension governance literature concerns investment performance, financial accounting and actuarial reporting details, based on annual financial reporting statements that do not include investment-level data. The unique role of public sector institutional investors ads an additional layer to the governance issues (Mallin 2012). Despite a fairly robust theoretical understanding of how agency problems might impact public pension fund management and outcomes, empirical studies of US public pension system structures remain inconclusive (see Mitchell 2005;Romano 1993 for a review). Romano (1993), Useem and Mitchell (2000) and Harper (2008) offer a stark example of how inconclusive and contradictory results have been, with results ranging from negative, insignificant and positive correlation between political trustees and fund performance, respectively.

Public pension fund investment
U.S. state public employee retirement systems cover almost 30 million state and local government employees with retirement and disability benefits and hold over $5 trillion in assets (NASRA, 2023). State retirement systems account for substantial portions of state budgeting planning, and annual obligations can amount to 5-15% of jurisdiction deficits. The portfolio allocations of the funds, therefore, attract the interest of many of the stakeholders described above, in addition to the investment community. The first significant hedge fund investment by a state public employee retirement system fund occurred in 2002, when the California Public Employees Retirement System (CalPERS) invested $50 million across 5 hedge funds (Rose-Smith 2011). By the time CalPERS exited their entire hedge fund portfolio in 2014, their investments totaled $4 billion across 24 hedge funds.
The period immediately following the 2008-2009 financial crisis was critical to state pension funds and alternative "hedged" investments. Between 2006 and 2012, the funds doubled their allocations to "alternative investments" and these moneys constituted about a quarter of total assets (Cliffwater 2022). Following the large declines in asset values in 2008-2010, many funds recovered with high returns as a result of some of these high yield investment choices. While alternative investments may offer benefits such as diversification and hedging for large investment portfolios, they also frequently come with high management fees and raise concerns about governance structures for monitoring externally managed assets. Moreover, even small changes within the class of alternative investments can have a significant impact on fund portfolios, and public pension funds are divided on their allocation to hedge funds and the percentage of management fees they pay.
Public pension funds invest in alternative assets for practical and theoretical reasons, including financial optimization and budgeting or accounting structures (Megginson et al. 2021). Alternative investments offer diversification and decreased correlation with the broader market for long-term fiduciary funds, but come with unique risks such as illiquidity and higher investment management costs. In recent years, both public and private pension funds have increased their investments in equities and alternative assets as part of their long-term portfolio strategies. In contrast with portfolios that gravitate toward less risky allocations as liabilities come due, stateowned investments have shifted to riskier allocations (Benzoni et al. 2007;Lucas andZeldes 2006, 2009;Andonov et al. 2012). One study found that in the period leading up to and including the 2008 financial crisis, public pension funds increased their portfolio risks following poor investment performance, suggesting a departure from typical portfolio maximization management (Benzoni et al. 2007;Lucas andZeldes 2006, 2009;Andonov et al. 2012).
Accounting rules may also play a role in portfolio allocations. Following a 2014 rule change from the Government Accounting Standards Board (GASB), public pensions were more restricted in their choice of interest rates used to discount future liabilities. Despite that change, a divide persists between academics and practitioner standards for discount rates (Brown 2009). Generally, economists consider the riskiness of the liabilities when determining the rate, rather than the expected growth of the dedicated asset funds (Novy-Marx 2009). Considering the protections many promised retirement benefits enjoy and the relative safety of future earned benefits, pension fund liabilities will not likely experience significant variation. It follows that the appropriate discount rate is akin to the risk-free interest rate. Since this rate is considerably lower than 8%, the newly calculated liability is significantly higher than financial statements would suggest. Linking liability discount rates with investment yields encourages risky investment strategies which have a higher average return but greater variability. Thus, public pension funds may pursue riskier strategies by virtue of actuarial realities.
Public pension funds run by state officials, despite the commitment of providing benefits promised to taxpayers efficiently, may allocate their investments different than other institutional investors. One study of state-owned investments found that large national sovereign wealth funds and public pension funds were far more likely to invest in private equity funds than other institutions (Megginson et al. 2021). Perhaps relatedly, public pension funds underperform compared to private investment funds, with some studies pointing to their poor investment choices (Bernstein et al. 2013;Andonov 2012). This paper does not compare overall performance of public and private pension funds; rather we attempt to evaluate investment choices in hedge funds by public and private pension funds in terms of diversity of investment within the funds themselves. Previous work looking at networks of Italian private pension fund investment find high degrees of overlap between funds (D'Arcangelis 2019). This relative lack of variety may increase correlated risk across different pension funds within the system. While heterogeneity can help minimize systemic investment risk by spreading it across industry and firms, specialization can also help reducing such risk through access to shared information, networks, and expertise in deal making (Bartkus and Kabir Hassan 2009;Norton and Tenenbaum 1993;Ro and Ziobrowski 2011).
The literature cited above suggests that public pension funds may pursue riskier portfolio strategies than their private counterparts for actuarial and management reasons. This study examines public pension fund investments in hedge funds by characterizing the network of investment flows and looks for evidence of similarity between pension fund type.

Data and methodology
We performed social network analysis (SNA) to explore dynamic networks between hedge funds and pension funds. In particular, we were interested in discerning how herd behavior as well as systemic risks differ between public and private pension funds. In order to examine the risk from the diversification perspective, we also performed a fixed effects regression analysis. In our knowledge, this is a novel attempt to explore the dynamic relationship between pension funds and hedge funds in a systematic fashion using micro-level investment data.

Data
The dataset on hedge funds and their investors was given to the authors by an exhedge fund manager. The dataset contained a wealth of information regarding investment in hedge funds between 2008 and 2012. This period immediately surrounding the financial crises and slow recovery during the Great Recession is a critical one for public pension fund investment. State pension plans lost an average of 25% of their asset value in 2008, and hedge funds mitigated some of the equity losses for fund portfolios that had placed significant investments (Pennacchi 2011;Rose-Smith 2011). The years immediately following the collapse in equity markets saw increased investment in hedge funds, according to survey data of the systems, accelerating a trend that had begun earlier (Rose-Smith 2011).
The dataset incorporates over 10,000 off-shore hedge funds, recording the fund names, the investors, as well as names and positions of nearly 2000 individuals on the boards of 5000 funds compiled from Security Exchange Commission (SEC) and other regulatory filings. It also includes the names of some institutional investors in these funds, which were compiled from U.S. Department of Labor filings, encompassing state, city, county and corporate pension funds, as well as university endowments and foundations. We analyzed the data between 2009 and 2011 as the data for 2008 and 2012 had a substantially smaller number of pension funds (33 and 32, respectively). For other years, the numbers of pension funds included in the data were 179, 202, and 205, respectively. The data only include defined benefit public pension plans.

Methodology
The SNA was used to visualize and analyze the investment patterns between public and private pension funds. For the network construction, the standard two-mode networks were projected into one-mode networks, i.e., the (two-mode) networks depicting investment from each pension fund to hedge funds were converted to the (one-mode) networks showing pension funds only, where each edge (i.e., link) representing co-investment by the linked pension funds to the same hedge fund (Fig. 1). Accordingly, a pension fund was removed from the networks if it invested in hedge funds but no other pension funds invested in those hedge funds. The advantage of the conversion is the ability to perform calculations of various network properties by only considering one dimension (i.e., pension funds) of the networks. For instance, when calculating a degree centrality score of the networks, the denominator of the score would be the total number of pension funds in the one-mode networks as opposed to the total number of pension funds and hedge funds in the two-mode networks. We explored both weighted and unweighted networks. Unweighted networks count the co-investment by two pension funds in a binary fashion (1 if there is at least 1 co-investment, 0 otherwise) while weighted networks count the number of co-invested funds between two pension funds. It should note that the network data established in this study are partially observed investments between pension funds and hedge funds. The missing data bias is likely to occur but unavoidable because the network imputation methods can only be applied to networks with random missing data (Smith 2022). For networks with non-random missing data, missing more central nodes generally yields a greater bias (Smith et al. 2017). While we only have data for 33 public pension fund systems in the United States, these data cover the vast majority of hedge fund investments by public funds. While exact public fund investment into hedge funds in unknown, data from annual performance surveys by Cliffwater found that 52 of 96 plans had a total of $63 billion invested in hedge For the network analysis, we evaluated network density and degree centrality. Network density is obtained for the entire networks by calculating the number of connections divided by the number of possible connections in the networks. It thus measures how well pension funds are connected with each other through common hedge fund investment. Degree centrality is obtained for each pension fund, measuring the degree (i.e., the number) of common investment that a pension fund has with other pension funds. Both network density and degree centrality were measured for weighted and unweighted networks. Conventional statistical tests including t-test for continuous variables and Chi-square tests for categorical variables were performed to examine the differences in investment patterns between private and public pension funds. In addition, the multi-level optimization of modularity was used for community detection to classify similarities of the pension funds' connections (Blondel et al. 2008). Here, each pension fund was assigned to a specific community through an iterative process where, in each iteration, funds are reassigned to the communities to which they make the maximum modularity contribution. Thus, the pension funds assigned to the same community have a similar hedge fund portfolio. All descriptive statistical analyses of networks were implemented using the programming language R (v4.0.5). For the visualization of the networks, both one-and two-mode networks were constructed for each year using Gephi (v0.9.2). Two-mode networks were constructed for both weighed and unweighted degree measures, while one-mode networks were prepared for unweighted degree only. This is because the amount of co-investment between connected pension funds differs in one-mode networks, and thus it is hard to define the appropriate measure for the degree of co-investment in hedge funds between two connected pension funds.
In order to examine the systemic risk taken by public and private pension funds, we calculated a diversification value for each type of pension funds based on Eq. (1): where investment ijt denotes the share of investment from pension fund i to hedge fund j in year t. The index is normalized to range between 0 and 1. Higher diversification index value represents the higher diversified investment portfolio while a lower value in the index represents less diversity (Alesina et al. 1999;Siddique 2021). In order to statistically assess the difference in the level of investment diversity between public and private pension funds, we estimated the ordinary square regression Eq. (2) with the community and the year fixed effect: where Y it represents the level of investment diversity by pension fund i in year t. public_pension_fund it is a binary variable taking the value 1 if pension fund i is public, and 0 otherwise. 1 thus captures the effect of the fund being public. s and t (1) are the community and year FEs, respectively. it is a random error term capturing all the unmodeled components influencing the level of investment diversity.

Descriptive statistics
Between 2009 and 2011, the total number of pension funds that invested in hedge funds was 235. Of those, 192 were private pension funds investing in 1976 hedge funds while 43 were public funds investing in 706 hedge funds. The breakdown of these numbers by year is shown in Table 1.
The average number and the median amount of hedge funds invested by each type of pension funds per year are summarized in Table 2. These numbers were consistently higher for public pension funds across years although we did not observe a significant difference in the numbers of hedge funds between private and public

Network analysis
The network involved 235 nodes (pension funds) for the study period (Table 3). The network was relatively sparse, yielding a density of 0.18. The average unweighted centrality of the nodes increased steadily during the study period (from 25.63 to 31.00 for private pension funds and from 22.66 to 30.38 for the public pension funds), indicating that an increasing number of hedge funds co-invested by multiple pension funds ( The average weighted centrality also increased steadily during the study period, and the amount of co-investment was significantly higher for public pension fund across the 3 years (all p < 0.005). Figure 2 visualizes the one-mode pension fund investment networks by year (Fig. 3, (1a)-(3a) for private pension funds and (1b)-(3b) for public pension funds). In the figures, red and green nodes represent private and public pension funds, respectively. The node size represents the degree of the node, i.e., the number of  . 2 One-mode unweighted network of co-investment in hedge funds by private and public pension funds (2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019)(2020)(2021) co-investments in hedge funds that the pension fund has with other pension funds. The red edges seen on the left side of Fig. 3 (1a)-(3a) reflect the investment in common hedge funds between two private pension funds while green edges seen on the right side panels of Fig. 3 (1b)-(3b) represent the investment in common hedge funds between two public pension funds. The blue edges seen on both sides of the figure represent the investment in common hedge funds between private and public pension fund. Finally, the number in each node represents the community that each node (pension fund) belongs to. The analysis detected up to 5 communities depending on the year and the type of funds.
The figures indicate that private pension funds are more likely to invest in the common hedge funds than public pension funds. This is demonstrated by the prevalence of red edges in comparison to blue edges in the panels (1a)-(3a). In contrast, the proportion of blue edges is significantly higher than that of green edges in the panels (1b)-(3b), indicating that, in the public pension fund network, the investment in common hedge funds between private and public pension funds (blue edges) is more common than the co-investment between public and public pension funds (green edges). This is also reflected in the relative abundance of red nodes (private pension funds) in panels (1b)-(3b) compared to the relative abundance of green nodes (public pension funds) compared to red in panels (1a)-(3a) across years. These observations together suggest that hedge funds in our data are likely to be invested by either: (i) private pension funds only; or (ii) both private and public pension funds, and hedge funds receiving investment only from public pension funds are relatively rare.
In terms of the density across the years, the figures indicate that the network density is comparatively low for 2010. It also appears that, for 2010, the distribution of co-investment is relatively skewed, showing a pair of private and public pension funds (shown in a thick blue edge) sharing a large number of common hedge fund investments and another pair of private and private pension funds (shown in a thick red edge) sharing a large number of common hedge fund investments. The former pair of private and public pension funds also appear in the 2011 networks. Table 5 summarizes the number and the proportion of the communities detected in the one-mode pension fund networks. Overall, the distribution of the communities differs statistically significantly between private and public pension funds (p < 0.001 for 2010, and p = 0.002 for 2011). For the periods that showed statistical significance, the community distribution was more skewed for public pension funds than for private pension funds, indicating a smaller number of communities within group. In particular, in 2011, more than 70% of the public pension funds belonged to Community 1, indicating that the investment pattern was relatively similar across these 70% of the pension funds, while the investment pattern was more distinct across private pension funds with 3 communities, each consisting of at least 25% of the pension funds. Similarly, in 2010, the majority (53%) of public pension funds belonged to Community 1, while the majority (75%) of the private pension funds were evenly split into two communities (Communities 1 and 4), each accommodating about 37% of the private pension funds.
Overall, results of the SNA analysis indicate that the public pension funds are investing more heavily in hedge funds in terms of both numbers and amount Community 4, n (%) 64 (37.65%) 2 (6.25%) Community 5, n (%) 4 (4.5%) -compared to the private hedge funds. Moreover, the investment pattern within public pension funds is more similar than that within private pension fund. The SNA analysis, however, does not shed light on the systemic risk taken by the public and private pension funds. To examine this aspect of the investment, we performed the regression analysis to test whether the hedge fund investment diversity differs between the two types of pension funds. Figure 3 shows the histograms of diversity index between public and private pension funds. The comparison of the histograms reveals that public pension funds, in general, have a higher level of diversity index compared to private pension funds. Table 6 reports the regression estimates. While the column (1.1) reports the effect of being a public pension fund in a bivariate analysis, the columns (1.2) and (1.3) report the estimates of the regressions after adjusting for community and year fixed effects, respectively. The column (1.4) presents the regression with both community and year fixed effects. The results confirm that public pension funds are more diverse in terms of investment to hedge funds than private pension funds (p ≤ 0.003). The coefficient indicates that the public pension funds are more diverse by about 0.01 on the diversity index than the private pension funds.

Discussion and conclusion
The public pension funds in our sample displayed significant homogeneity in their hedge fund investments, to a greater degree that the private pension funds in our sample. While our results are suggestive of overlap between public pension funds in terms of hedge fund investment choice, we want to repeat caveats about our data. These data are incomplete, covering a non-random sample of public and private pension fund investments into hedge funds. We do not know the degree of omitted investments, and, therefore, must interpret network level characteristics cautiously. That said, we are confident that among the funds that are included in our data, there is not systematic selection of investment reporting, and in aggregate, our data cover a significant part of total public pension fund investment in hedge funds in the years studied.
The overlap in hedge fund choice is coupled with a variety of hedge fund investments for each public fund. On one hand, this is a surprising result, as we expected to see state-specific hedge fund investments to comply with (or preempt) state investment guidelines, leading to less overlap between state funds. On the other hand, the state funds may have included common compliance standards and/or paid consulting paths to find appropriate hedge funds, resulting in the overlap. We acknowledge that this finding may be biased due to the confounding factor of fund size, which may be influencing both pension fund type and diversity. While the fund size data for public and private pension funds are available from the Center for Retirement Research at Boston College and the US Department of Labor, respectively, most of these data are aggregated at the state level, and thus could not be merged sufficiently to our fund level data.
A related explanation for the relatively strong overlap of hedge fund choices involves the communities of practice among public pension fund chief investment officers. Chief Investment Officers for public pension funds would meet periodically at conferences and workshops to discuss trends in investment, stakeholder needs and implementation. As public funds increased their investments into alternatives, they were faced with a strict fee structure that was seemingly non-negotiable in contrast to traditional equity funds. The fee structure varied by alternative investment type, but was often described as a "2 and 20," referring to a flat management fee of 2% for assets plus 20% of profits generated by the investment. This fee structure for hedge funds and other alternative investments is often the source of ambiguity regarding public pension fund investment performance, since the high fees can eat into excess returns but variation in reporting standards impede clean comparisons. To make matters even more complicated, many public pension funds invested in "funds of funds," wherein a manager provides access to multiple hedge funds and/or private equity funds and subsequent diversification. The fee structure thus became a layered 2-and-20 plus some spread for the parent fund.
Public pension funds, while large, generally spread their investments around and, therefore, generally do not always have the ability to invest directly in vehicles with lower fees, nor do they have leverage to negotiate the 2-and-20 fees. The community of practice among public pension funds investment officers, however, may help disparate funds identify favorable hedge funds. While we do not study personal communication networks among investment officers here (a notoriously difficult prospect, as described in Ahern 2017), we do note documented cases in which overlapping investment choices are explicitly designed. In some cases, smaller public pension funds are known to have collaborated to create a request for proposals, in which they invite alternative investments to bid for their investment assets via lower fees (for example, see comments by then-CIO at the Orange County Employees Retirement System in (Lernov 2014). This type of RFP would definitionally result in greater homogeneity among public pension fund as a matter of strategy.
This analysis provides a novel perspective of the rapid shift of public pension fund assets into hedge funds and alternative investments, although it does not address the fundamental ambiguity regarding overall fund performance. The flow of public money into hedge funds creates communities of potential influence on portfolio outcomes and allocation decisions, and there is value in understanding the concentration of institutional investment by public pension funds, even with incomplete data. Considering the propensity during the period under study to use fund of funds for alternative investment, there is potential for intricate networks at the individual, rather than institutional level as well. Consider a placement agent or manager that creates the fund vehicle for multiple public pension funds, while the public pension funds' assets would be recorded in these data as an investment into a hedge fund, the pathway for this investment is through a particular agent. The degree of concentration in the network of agents is unclear in this analysis and would be worthy of study in the future.
In this paper, we attempt to evaluate investment choices into hedge funds by public pension funds by comparing their homogeneity with private funds. While the limitations of our data preclude us from addressing the fundamental question of overall investment performance, our descriptive analysis indicates patterns of investment overlap across different public pension funds. This is significant because much of the pension investment literature emphasizes the "wide variation" in portfolio allocation strategy between public pension funds, when comparing broad investment categories such as equities, fixed income and alternative investments (see, for example, Megginson et al. 2021). Our study suggests that this wide variation may not carry through to distinct elements of portfolio investments, with an emphasis here on hedge funds within the alternative investment category. In other words, it is possible that, contingent on investing a threshold amount in hedge funds, there may be less variation in hedge fund investment vehicles among public pension funds than previously assumed.
Additional study is warranted here for two reasons. First, understanding microlevel portfolio allocation may inform the literature about overall fund strategy, especially in more opaque alternative investments. Second, the homogeneity may suggest the importance of shared information, networks and expertise in deal making in public fiduciary fund investment. Given the public budgeting implications of portfolio performance of public pension funds, the potential impact of these communities of practice deserve greater attention. This paper provides a snapshot of the direct investment networks during a critical period of portfolio volatility; the next steps clearly involve updating and filling out this picture for future research.
Author contributions All the authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by SL, M-HL, ABS, RK, and NK. The first draft of the manuscript was written by SL, M-HL, ABS and RK, and all the authors commented on previous versions of the manuscript. All the authors read and approved the final manuscript.
Funding Partial financial support was received from the Institute of New Economic Thinking.

Data availability
The datasets analyzed during the current study are available from the corresponding author on reasonable request.

Conflict of interest
On behalf of all the authors, the corresponding author states that there is no conflict of interest.

Research involving human participants and/or animals Not applicable.
Informed consent Not applicable.

Consent for publication
All the authors (1) made substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data; (2) drafted the work or revised it critically for important intellectual content; (3) approved the version to be published; and (4) agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.