Investment risk-taking and benefit adequacy under automatic balancing mechanism in the Japanese public pension system

The study addresses how investment risk-taking in Japan's public pension reserve fund affects benefit adequacy and long-term sustainability under the automatic balancing mechanism (ABM). Japan faces rapid population ageing, low fertility, and uncertain economic growth, making it difficult to simultaneously maintain sustainability and adequate benefits. The ABM (introduced in 2004) automatically adjusts benefits via modified indexation to restore solvency while premiums are capped. The government targets a replacement rate (RR) of at least 50% and plans for a reserve-to-expenditure ratio (RER) of 1 at the end of the verification period. However, because economic and financial markets fluctuate and ABM introduces non-linear benefit adjustments, scenario-based forecasts are insufficient to capture risk. The research question is how different stock weightings in the reserve fund influence the joint distribution and downside risks of RR and RER under stochastic economic and market conditions, and whether dynamic investment strategies can improve risk outcomes.

Prior work documents ABMs across countries (e.g., Sweden, Canada, Germany, Finland, Japan) and evaluates their effects on solvency and adequacy. Vidal-Meliá et al. (2009) argue ABMs and solvency indicators can bolster credibility and reduce political manipulation. Boado-Peñas et al. (2020) find trade-offs between sustainability and adequacy where higher equity allocation can raise RR within mixed PAYG-fund systems. In Japan, Fujisawa and Li (2012) assess ABM effects via stochastic mortality/fertility but without stochastic economics; Hizu (2020) studies RER without fund risk-taking; Nakashima and Kitamura (2021) examine RR with a simpler model. The paper extends this literature by jointly analyzing RR and RER distributions, focusing on equity risk-taking (the main controllable policy lever under Japan’s ABM), and by testing dynamic strategies (portfolio insurance and contrarian) within the ABM framework.

The authors implement the Ministry of Health, Labour and Welfare (MHLW) actuarial verification program (FORTRAN/C++) to project insured persons, pensioners, benefits, revenues, reserve fund evolution, and the ABM’s modified indexation over 2019–2115. Key stochastic variables are inflation, wage growth, and reserve fund investment returns; all are modeled as normally distributed with means calibrated to government scenarios and GPIF sources, and with historical standard deviations and correlations (primarily from 1993–2017 data). The pension system comprises the Basic Pension (BP) and Employees’ Pension Insurance (EPI). Benefits for both tiers and EPI notional accrual wages are adjusted annually by the modified indexation rate that depends on inflation, wage growth, demographic factors, and a constant parameter. The ABM applies until the projected RER equals 1 at the end of the verification horizon; when economic conditions are unfavorable, ABM can persist longer, lowering RR. RR is defined as benefits for a representative household relative to representative wages, and RER is the prior-year reserve fund divided by total expenditure. The reserve fund evolves via investment returns and net cash flow (revenues minus expenditures). Investment strategies: constant mix stock weights (SW) of 0%, 25%, 50%, 75%, 100%, with assets limited to domestic stocks/bonds and foreign stocks/bonds (benchmarked to TOPIX, Nomura-BPI, MSCI ACWI ex-Japan, FTSE WGBI ex-Japan/China). Economic assumptions follow MHLW’s 2019 actuarial valuation: a standard (Case 3) and a pessimistic (Case 5) scenario; the medium population projection is used. Monte Carlo: 9000 replications per SW and economic case. Robustness checks include (i) varying which variables are stochastic (returns only; wage only; inflation only), (ii) doubling/halving asset return volatilities, and (iii) polar allocations (foreign-only vs domestic-only) at SW=50%. Dynamic strategies: rule-based portfolio insurance (reduce SW as RER falls; increase as it rises) and a contrarian rule (increase SW as RER falls; decrease as it rises), with SW adjustments at 2040, 2060, 2080, and 2100 for paths where ABM has not stopped. Risk metrics include percentile paths, 10th percentile (VaR), and marginal CVaR (mean below VaR), as well as probabilities that RR or RER fall below their respective VaR.

- Under pessimistic economic assumptions (Case 5), both RR and RER exhibit substantial downside risk despite the ABM. The 10th percentile RER often falls below zero during the simulation, indicating possible depletion of reserve funds, while RR can fall well below the government’s 50% target; median RR values cluster around approximately 38–40% in later years for most SW, failing to reach 50% even in the standard case (Case 3). - High SW increases the upside (higher 90th percentile RR and RER) but deepens downside risk: for SW=50% under Case 5, the 10th percentile RER turns negative from roughly mid-century (e.g., around 2065–2106 the 10th percentile RER is negative and can reach about −0.5), before improving later as ABM continues to cut benefits. Similar or larger negative tails occur for SW=75% and 100% (e.g., 2075 RER 10th percentile near −0.79 and −1.33, respectively). For SW=0%, negative tails are smaller or later. - The timing of ABM termination is stochastic; in high-SW environments some paths never exit ABM within the horizon (e.g., basic pension non-termination shares around 20–26% for SW=50–100% in Case 5), causing persistent RR declines. Early ABM stops in higher SW paths elevate RR medians/uppers but leave RER to revert toward 1 later. - Joint-risk view: RR and RER VaR and CVaR show that moderate SW (25–50%) often balances risk and return best. At 2075 under Case 5, the 50th percentile RR and RER peak around SW=50–75%, while VaR/CVaR for both metrics are least adverse for lower SW (0–25%). The probability that RR or RER is below its VaR is relatively flat around SW=25–75%, while the probability both are below VaR rises with SW to about 6% by SW=50% and flattens thereafter. - Government obligation when RER becomes negative: the probability that RER is negative for at least one year is 20%. Conditional present values of obligations range from about JPY −9.5 to −13.0 trillion (cashflow-based measure) and JPY −11.8 to −65.3 trillion (balance-based measure), depending on discount rate. This equals approximately 0.9–6.4% of outstanding JGBs and up to about 11.4% of GDP (balance-based measure). The ABM’s benefit cuts mitigate cashflow-based obligations relative to the balance-based measure. - Robustness: Investment return volatility is the dominant driver of RR and RER variability; making only returns stochastic produces results close to the full-stochastic baseline, whereas stochastic wage or inflation alone has minor impact. Doubling return volatilities worsens downside metrics; halving improves them (with some non-monotonic effects due to ABM’s non-linearities). Foreign-only allocations (at SW=50%) raise upside but deepen downside RER risk (lower 10th percentile) relative to mixed domestic/foreign; domestic-only lowers expected outcomes. - Dynamic strategies: Portfolio insurance improves downside protection (higher VaR/CVaR for RR and RER versus constant SW=50%), especially in later years (e.g., 2100 RER VaR improves from −0.29 to +0.17; 2115 RER VaR from 0.67 to 1.00; 2115 RR VaR from 30.2% to 32.9%), but reduces medians. The contrarian rule increases median RR and RER relative to constant mix in several horizons but tends to worsen downside tails (lower RER VaR/CVaR than portfolio insurance).

The results confirm a fundamental trade-off under the ABM: higher equity exposure raises median and upper-tail adequacy (RR) and sustainability (RER) metrics, but also magnifies downside risks due to volatile investment returns. Because ABM reduces benefits until solvency is restored, protracted weak markets at higher SW can lock the system into persistent benefit cuts, depressing RR and delaying recovery of RER. Conversely, too little risk-taking (SW=0%) largely avoids deep negative tails but almost certainly leads to low RR locked at a reduced level due to insufficient returns. Thus, moderate risk-taking (roughly SW 25–50%, and sometimes up to 75% depending on valuation horizon) balances risk and return for joint RR–RER goals. Joint VaR/CVaR analyses show that ABM’s path-dependent termination amplifies non-linearities; RR and RER risks need to be managed together, not in isolation. Dynamic stock-weight rules can be useful policy levers: portfolio insurance helps contain downside risk (limiting probabilities of negative RER and improving VaR/CVaR), while contrarian tilts can enhance medians but at a risk cost. Policymakers may also contemplate temporary borrowing (negative RER) that is effectively repaid as ABM reduces benefits. Given the government’s RR target of 50% is rarely met at the median under the modeled assumptions, meeting adequacy goals may require explicit policy choices (e.g., contribution changes outside current ABM, or acceptance of higher investment risk with risk controls).

Using the official Japanese actuarial simulation framework with stochastic macro and market inputs, the study shows that both benefit adequacy (RR) and sustainability (RER) bear significant downside risk under the ABM. Extreme equity risk-taking increases tail risk of depleted reserves and persistent benefit reductions, while zero-risk strategies lock in low RR. Moderate stock weights generally provide the best balance between risk and return across horizons. Portfolio insurance rules can reduce downside risk in RR and RER relative to a constant mix at the cost of lower medians; contrarian rules raise medians but worsen tails. The government faces non-negligible contingent obligations when reserves are depleted, though ABM-induced benefit cuts moderate cashflow burdens. Future research should optimize SW dynamically under multi-objective criteria (joint RR–RER risk/return), incorporate default risk for JGBs, explore alternative assets and broader allocation mixes (domestic vs foreign), allow for time-varying volatilities and mean reversion, and evaluate policy tools beyond SW (e.g., contributions, retirement age) within legal constraints.

Key limitations include: (1) No modeling of Japanese sovereign default risk despite large JGB holdings; (2) Only the medium variant of population projections is used; (3) Investment universe restricted to four traditional asset classes, excluding alternatives; (4) Normality assumption for stochastic variables with means set by government scenarios and constant volatilities/correlations (except robustness), potentially understating tail dynamics; (5) Investment returns modeled as random walks (no mean reversion or regime shifts), possibly affecting evaluation of dynamic strategies; (6) ABM rules are applied as in law, with discrete evaluation of dynamic SW rules (every 20 years), so portfolio insurance floors may not hold each year; (7) Government policy parameters other than SW (contribution rate, retirement age, indexation rules) are fixed; (8) Some results are sensitive to return volatility levels and domestic/foreign splits, suggesting model risk.