An Application to U.S. Bank “Available for Sale” Portfolios
Historically, banks have generated a material percentage of their net interest from loan activities. Since the 2008 crisis, the lack of loan demand and the need to rebuild liquidity and comply with proposed new Basel III requirements have caused U.S. bank securities portfolios to grow by roughly $1 trillion, nearly doubling in size and making them a substantial source of overall bank profitability. Additionally, because of the proposed Basel III capital rules, banks' capital requirements may be heavily influenced by the unrealized gains and losses within the available for sale (AFS) portion of the securities portfolio, including the performance of the AFS portfolio under supervisory stress scenarios.
Given the growing size of bank investment portfolios as a percentage of average earning assets, evolving supervisory requirements and the current low rate environment, banks are revisiting their asset allocation strategies within their investment portfolios to try to meet several competing objectives. Bank portfolio managers aim to prudently optimize the yield on their portfolio while following applicable internal investment guidelines. At the same time, they need to ensure that a certain portion of the portfolio is invested in relatively liquid assets, ensure the tail risk properties of the portfolio are in check, and that the asset mix does not generate additional regulatory capital requirements beyond the institution’s tolerable limits. Finally, portfolio managers have to be cognizant of the performance of their portfolios under both internally specified and external/supervisory stress scenarios.
Current AFS portfolio allocations are highly concentrated in U.S. Treasury securities and Agency mortgages and as such have substantial duration risk. As AFS portfolios have grown, banks have generally invested in relatively low-risk and liquid assets. We believe banking institutions of all sizes can notably increase risk-adjusted returns within their AFS portfolios despite these supervisory and institutional constraints through modest reallocations. Currently, AFS allocations to Agency mortgages may be too large, while credit, non-U.S. corporate bonds and select structured securities can be attractive for banks of all sizes. Moreover, certain banks might benefit from small increases in high yield exposure as well.
Asset allocation for U.S. banks: A case study
Recognizing the complex portfolio optimization/asset allocation needs of the U.S. banks, PIMCO has developed an analytical framework that integrates four key components (see Figure 1) to provide customized solution analyses that:
- Utilize PIMCO’s forward-looking capital markets assumptions (CMAs), including its asset-class-level implications on returns1
- Model banks’ internal and external constraints, such as risk-weighted assets (RWAs), investment guidelines, asset allocation and liquidity constraints
- Impose multi-period, multivariate scenarios (such as risk Federal Reserve's CCAR / Dodd-Frank Act Stress Test [DFAST] scenarios) to assess portfolio performance over time
- Suggest allocations that can potentially minimize tail risk in the portfolio per unit of return
First, we review an asset allocation analysis on benchmark U.S. bank AFS portfolios in the presence of sample investment guidelines, liquidity constraints, risk-weighted asset constraints, and performance constraints in key macroeconomic scenarios, and assess the resulting asset allocation outcomes across U.S. banks.
1. Assumptions regarding future returns and the investment universe
A key input to the optimization process is the set of forward-looking risk and return assumptions across various asset classes. PIMCO has designed a five-step framework to produce its capital market assumptions, or estimates of future asset-class-level returns, that combines empirical analysis, model-based results and qualitative surveys by investment professionals. The combination of qualitative assessment with quantitative grounding and review is designed to harness expertise regarding market forces and features that may be difficult to model quantitatively, yet still produce sound, internally consistent estimates across various forecasters and asset classes.
Figure 2 depicts the five steps in the PIMCO framework used to generate forward-looking return assumptions. Using this, or other processes as determined by the client, portfolios can be characterized by combinations of asset-class benchmarks and indexes, each with their own combinations of risk factor loadings and return assumptions. Asset allocations within bank AFS portfolios are represented in the below sample analysis by common asset-class benchmarks.
As Table 1 illustrates, the estimated returns are consistent with PIMCO’s outlook for a New Normal economy with returns for most asset classes expected to be well below their historical averages.
This is particularly true for fixed income asset classes, as rates and spreads are near historically low levels limiting both returns and the potential for future capital appreciation. For U.S. Treasuries, the secular real policy rate is expected to remain slightly negative for the next few years. U.S. Agency mortgage-backed securities (MBS) securities are heavily influenced by Federal Reserve activity in the market, and we expect their duration-matched returns to only slightly exceed Treasuries. While credit spreads seem to be relatively tighter than historical averages, they are not at the tightest levels historically. Emerging market (EM) economies are expected to continue to outpace the developed world over the secular horizon, though their path of expansion and transition will be more volatile. Finally, equities are expected to see performance better than the last decade, but worse than the long-term averages as demographics generate a headwind against outsized returns.
2. Bank AFS portfolios
In this example we model bank portfolios using asset-class- level weights on benchmark indexes that are representative of typical AFS portfolio asset allocations, where the asset-class portfolio weights are calculated by the PIMCO Financial Institutions Group. The bank portfolio data contains asset- class allocation weights for banks of five different sizes: banks with total assets less than $1 billion, between $1 billion and $10 billion, between $10 billion and $50 billion, between $50 billion and $250 billion, and, finally, larger than $250 billion in total assets.
Table 2 provides details regarding the asset allocations of bank AFS portfolios by institution size, as well as the implied portfolio-level estimated returns, volatility, and 99th percentile conditional value at risk (CVaR). In the table liquidity assets include cash and deposits, U.S. Treasuries, Agencies and Agency MBS; spread products include investment grade (IG) and high yield (HY) corporate credit, and structured finance instruments; international assets include exposures in the developed and emerging markets; and municipal and equity cover municipal bonds, and equity/mutual funds and equity strips respectively.
As Table 2 illustrates, typical U.S. bank AFS portfolios are quite concentrated in cash, government and Agency products, reflecting the emphasis on liquidity and self-insurance against tail risks. Estimated portfolio returns seem to increase nearly monotonically with total asset size, while risk measures seem to follow a slightly different pattern. Both total portfolio volatility and tail risk (99th percentile CVaR) initially decrease with institution size only to increase again amongst the very largest banks. This appears to be driven by an initial decrease in municipal allocations in favor of IG credit as we move from the smallest banks to the mid-sized organizations, and then the trend is reversed as larger bank allocations shift more resources into high yield, asset-backed securities and international corporates. Overall though, institutions of all sizes appear to hold substantively similar portfolios.
3. Tail risk and mean-CVaR optimization
Typical portfolio optimization routines generate “mean- variance efficient” portfolio allocations. However, given the AFS portfolio’s emphasis on income and providing a source of liquidity through the cycle, we believe that a focus on the second moment is inappropriate in this context, and that a more relevant focus is on the portfolio performance far into the tail of the loss distribution. To this end, we calculate a series of “mean–tail risk efficient” portfolios that seek to minimize tail losses as defined by the CVaR at the 99th percentile. As a result, the allocations generated by this optimization routine can be readily understood in the context of liquidity provision in extreme loss scenarios. The mean–tail risk efficient portfolio given a target level of return can be represented by the following optimization:
w is a vector of portfolio allocations across asset classes
r is the realized portfolio return
α is the percentile identifying the tail (99% in this analysis)
f(w,r) is the portfolio performance
VaRα(w) is the Value at Risk given allocation w and tail percentile α
[µ] is a vector of return assumptions for each asset class
This process is repeated for each candidate return (r) to determine the frontier of efficient mean–tail risk portfolios. In short, this specification alters the objective function so that the recommended portfolio allocations minimize CVaR rather than simply minimize variance.
Liquidity and investment guidelines
As AFS portfolios are designed to provide liquidity throughout the economic cycle, we first place restrictions on the portfolio allocations to help guarantee a certain level of liquidity. We require AFS portfolios to have a minimum of 60% allocation to low-risk, liquid securities, such as cash (and cash equivalents), U.S. government and U.S. Agency securities. We also require the remainder of the portfolio allocation to be in a reasonable neighborhood around observable AFS portfolio allocations and retain some minimal level of diversification: Total allocations to any non-liquid product cannot exceed 15% of the total portfolio, except in the case of international bonds at 10%, and equities, where the maximum tolerable allocation is reduced to 5%.
Risk-weighted asset and scenario performance constraints
In addition to internal constraints on the overall asset allocation, there are several supervisory requirements that banks must monitor and manage. Among the most common are the risk-weighted asset (RWA) charges, which contribute to regulatory capital requirements. Bank AFS portfolio allocations must incorporate restrictions on admissible RWA levels. Accordingly, we estimate the current bank portfolio’s RWAs by calculating the risk-weighted asset total for each asset-class benchmark using the standardized approach to credit risk under Basel II. We combine the asset class level RWA estimates with the bank portfolio allocations to estimate total RWA for the typical bank AFS portfolio, and then constrain optimal AFS portfolio allocations to have RWA totals no greater than their current estimated value.
Stress and scenario-based performance requirements are another key constraint on bank AFS portfolio allocations. In the aftermath of the 2008 crisis, stress tests and scenario evaluations became an essential analytical tool for portfolio and risk managers in the banking industry. We expect scenario testing to become still more critical with the adoption of Basel III rules. In the current regulatory regime, provided that the bank’s capital plan passes supervisory stress tests, stress outcomes do not have a direct effect on the bank’s asset allocation decisions. However, with the onset of Basel III, stress tests are poised to directly affect capital requirements: Banks will be required to hold a larger portion of their total capital in the form of common equity with the creation of a new common equity Tier 1 (CET1) ratio. In calculating the new CET1 ratio, banks would be required to include accumulated other comprehensive income (AOCI) as part of CET1, where for many banks the primary driver of AOCI is unrealized gains and losses in the bank’s AFS securities portfolio. The current proposals include accounting for unrealized gains in regulatory capital calculations, e.g., via losses incurred in regulatory stress testing, by the removal of the AOCI filters from regulatory calculations, such as the Dodd-Frank Act Stress Testing program. At the same time, supervisory stress testing will be required of many more institutions: Among its many provisions, the Dodd-Frank Act (DFA) will expand the scope and breadth of stress testing practically to the entire banking system (to all banks with assets over $500 million). Needless to say, the adoption of Basel III and bank portfolio scenario testing programs, such as the Dodd-Frank Act Stress Testing program – with the AOCI filter turned off – are likely to contribute to unrealized losses in AFS portfolios, which in turn will translate to higher regulatory capital for U.S. banks. For these reasons, one can anticipate that the introduction of Basel III/DFA rules will further formalize and strengthen the existing links between stress scenario performance of bank portfolios and economic incentives of the banks in portfolio asset allocation.
Recognizing these emerging trends, PIMCO has expanded its capabilities to include an assessment of the sensitivities of any portfolio to key international macroeconomic and financial indexes. The analytical model covers 26 economic/financial variables and can be utilized to capture macroeconomic scenarios, such as those within the Federal Reserve’s Comprehensive Capital Analysis and Review (CCAR) program. The model translates shocks to supervisory macroeconomic variables into shocks to factors that more directly affect asset pricing and risk analysis through principal components analysis and multi-dimensional projection.
To characterize the importance of stress scenario performance on admissible portfolio allocations, we include in the optimization process performance-based constraints under a widely known external macroeconomic scenario: the Federal Reserve Bank’s “severely adverse” scenario from the 2013 Comprehensive Capital Analysis and Review of U.S. banks (please see Table A1 in Appendix for more information regarding the stress scenario).
Table 3 contains the restrictions on AFS portfolio liquidity, investment allocation constraints, risk-weighted asset charges and scenario performance used in this analysis by asset class.
In this section, we present the findings of the optimization problem described in the previous section. First, we consider the case where banks are only subject to the liquidity and investment guideline constraints without any supervisory restrictions. The resulting efficient frontier is portrayed in blue in Figure 3. Next, we incorporate the RWA charges and scenario-based performance constraints. These constraints shift the efficient frontier inward as displayed in gold in the same figure.
Based on Figure 3 (and Table 1) one observes that while there are some nuances and size-specific behavior, overall the portfolio compositions and risk-return profiles across banks of various sizes are quite similar. This is of import: It suggests that external asset-specific shocks, such as regulatory rules which provide asset-specific incentives/disincentives that shift the efficient frontier – may trigger reactions across the industry en masse.
Current allocations of AFS portfolios are located strictly within the interior of the efficient frontier. In the presence of the liquidity and allocation constraints alone, banks can markedly improve their portfolio returns while holding tail risk constant. In this sample analysis, these gains range from 44 bps – 59 bps depending on the size of the institution. Increases in returns of this magnitude would roughly double the estimated returns on AFS portfolios. Mid-size banks observe the smallest gains from the optimization, though the efficient frontier still attains a substantial increase in expected total returns relative to their current allocation. In terms of portfolio allocation, the optimization subject to liquidity and allocation constraints suggests a reduction in Agency MBS, and an increase in short duration credit, emerging market (EM) and equities exposures for banks of every size. The smallest and largest banks would benefit from a modest increase in high yield exposure as well. The detailed optimal allocations with liquidity and allocation constraints alone are included in Table A2 in the Appendix.
The addition of RWA and scenario performance constraints shifts the efficient frontier inward for the highest-return, highest-tail-risk allocations. Given current CVaR levels of bank portfolios, the introduction of the regulatory RWA and scenario constraints reduces the attainable level of returns relative to those that would be possible in the presence of only liquidity and allocation constraints. These reductions are the largest for the smallest banks in our sample analysis (19.9% in relative terms, or a 9 bps decline in absolute returns) and the largest banks (12.9%, 7 bps), where medium sized banks are affected relatively less (2.8%–4.6%, 1 bps–2 bps). The reduction in attainable tail-risk-adjusted returns can be thought of as the implicit cost of RWA and scenario requirements on bank AFS portfolios. Given the current size of these portfolios, such a reduction in returns would imply an industry-wide cost ranging from $750 million to $4.5 billion per year. The detailed portfolio optimization results that incorporate liquidity, allocation, RWA and scenario performance constraints can be found in Table A3 in the Appendix.
In terms of portfolio allocation across asset classes, the introduction of the RWA and scenario constraints does not alter the direction of asset allocation recommendations vis-à-vis the current portfolio compositions and the unconstrained frontier. The supervisory-constrained optimization suggests a reduction in Agency MBS and an increase in short duration credit, emerging markets (EM) exposure and equities across banks of all sizes. Similarly, we see an increase in high yield exposures for small banks and a decrease in ABS allocations for the largest institutions, though overall the recommended reallocations are extremely similar across institution size.
To summarize the impact of the different constraints, we have calculated the equal-weighted arithmetic average optimal portfolio across the five different size-specific portfolios utilizing the five asset groupings: liquidity, municipals, spread instruments, international exposure and equities. Figure 4 below illustrates the calculated average asset-class-specific distributions across banks of various sizes for three portfolios: the average current AFS portfolio; the average liquidity and allocation constrained optimal portfolio; and finally, the average liquidity, allocation, RWA, and scenario constrained optimal portfolio.
Figure 4 illustrates three key points: 1) impositions of constraints appear to have non-negligible asset allocation implications on AFS portfolios and 2) the optimization results suggest that AFS portfolios would benefit from a reallocation away from liquidity products, primarily in Agency mortgage holdings, and into spread (particularly short duration corporate credit), international (EM), and equity classes to the extent permitted by their liquidity constraints and institutional asset allocation guidelines, and 3) the further imposition of supervisory constraints makes certain spread products slightly less attractive in terms of controlling for tail-losses in the portfolio.
Our findings suggest that banks can markedly improve the performance of their AFS portfolio while holding their tail risk and regulatory metrics constant through modest portfolio reallocations. We find that:
- Imposition of the internal/external constraints (such as liquidity, investment guidelines, RWAs, CCAR or other scenarios) appear to have non-negligible asset allocation implications on AFS portfolios
- U.S. bank AFS portfolios may benefit from a reduction in Agency MBS exposure and an increase in short duration corporate credit, along with small increases in equity and emerging market holdings
- The introduction of supervisory requirements on the sample portfolio above and beyond internal allocation constraints reduces attainable returns by 3%–19% (1 bp–9 bps), costing the banking industry between $750 million and $4.5 billion annually in forgone returns within the AFS portfolio
- As the anticipated removal of the AOCI filter under Basel III will cause unrealized losses to flow through to regulatory capital calculations, regulatory stress scenarios and assorted supervisory restrictions are poised to have a larger impact on bank portfolio asset allocations
- Finally, as banks across different sizes tend to hold similar assets in fairly similar allocations within their securities portfolios, the asset-class incentives and disincentives embedded within supervisory scenarios, combined with their direct effect on capital under Basel III, may induce market-wide reallocations across asset classes
1 PIMCO custom client reports are created in consultation with the client and may be based upon client, consultant, index or PIMCO capital market assumptions. Note: Capital market assumptions are derived from proprietary research and modeling and are not a guarantee of future results.