Understanding HANK: Insights From a PRANK

Households have CARA utility, face idiosyncratic labor-endowment shocks that leave individual income normally distributed around the aggregate mean, and can save in a single riskless bond, without imposing the zero-net-liquidity limit that most other tractable HANK models rely on (Section 2, pp. 3-8). The authors emphasize that avoiding the zero-liquidity limit is what allows a genuine precautionary-savings channel of monetary policy to appear – “we do not rely on the zero liquidity limit… This allowed us to uncover an additional dimension through which monetary policy affects aggregate demand” (Conclusion, p. 30) – because with strictly positive liquidity, consumption risk and income risk are no longer trivially identical, and the common marginal propensity to consume μ_t becomes a variable that monetary policy itself can move.

Aggregating individual incomes shows that the variance of idiosyncratic income, σ²(Y), and its cyclicality dσ²(Y)/dY, depend on four factors: the cyclicality of real wages, of labor tax rates, of firms’ dividend policy (how unequally dividends are distributed in booms versus recessions), and of employment/endowment risk (Section 3.4, eq. 18-19, pp. 11-12). Critically, “just as the level of taxation affects the level of risk, the cyclicality of fiscal policy affects the cyclicality of risk” (p. 12): a labor tax rate that falls in recessions and rises in booms tends to make risk countercyclical, while lump-sum transfers concentrated in recessions and financed by proportional taxes tend to make risk procyclical – so ordinary automatic-stabilizer design directly governs which regime the economy is in.

Proposition 3 derives the exact determinacy condition Φ_π > 1 + (γ/κ)²(1-β̃)/(1-β̃) + γβ̃Λ, which collapses to the standard Taylor principle Φ_π>1 only when risk is acyclical (Θ=1) (Section 4, eq. 25, pp. 14-15). With procyclical risk (Θ<1) the required coefficient falls, and “if risk is sufficiently procyclical, then determinacy is guaranteed even under a nominal interest rate peg, contrary to the classic result of Sargent and Wallace (1975)” – consistent with Auclert et al.’s (2017) numerical finding that HANK models can be determinate at lower Taylor coefficients. With countercyclical risk (Θ>1), “the standard Taylor principle… is not even sufficient for determinacy,” which the authors show is exactly the case studied numerically by Ravn and Sterk (2017): “indeterminacy is more likely in HANK models only when income risk is countercyclical,” not as a general property of incomplete markets (Section 4, p. 15).

Iterating the HANK IS equation forward under fixed prices shows the sensitivity of date-t output to a rate cut announced for date t+k scales with Θ^k; when Θ>1 (countercyclical risk) this sensitivity is increasing in the horizon k, replicating and amplifying the forward guidance puzzle, whereas when Θ<1 (procyclical risk) it is decreasing in k (Section 5.1, eq. 26-27, pp. 17-19). The amplification channel under countercyclical risk works because higher expected future output at t+k also lowers expected future risk, so households at t+k-1 raise spending more than one-for-one, “leading to a larger boom at date t+k-1,” which cascades backward through time; a second channel operates because lower future real rates raise μ (the MPC), making consumption more responsive to income and reinforcing the boom (p. 18). The authors are explicit that this is not driven by constraints or myopia: “our households are infinitely lived and unconstrained… it is because idiosyncratic income risk is procyclical [or countercyclical]” (p. 18).

No, but its effects are shown to be logically distinct from – and, for determinacy specifically, entirely absent relative to – the effects of risk cyclicality. Introducing a fraction η of HTM agents whose average income moves with aggregate income according to y^HTM_t=χy_t (following Bilbiie 2008, 2017), the paper proves that at χ=1 “aggregate outcomes are unaffected by the introduction of HTM households” regardless of η (Section 6, pp. 22-23, fn. 37); away from χ=1, MPC heterogeneity “makes a discounted Euler equation ’less discounted,’ and an explosive Euler equation ’less explosive’” when χ<1, or amplifies the deviation from RANK when χ>1 – but “it cannot turn a discounted Euler equation into an explosive one, or vice versa” (Section 6.2, p. 25). For determinacy specifically, Proposition 3’s condition is shown to be entirely unchanged by the introduction of MPC heterogeneity: “the cyclicality of income risk remains the most important feature affecting determinacy” (p. 26).

Unlike for determinacy, MPC heterogeneity does interact with the forward guidance puzzle: if risk is procyclical enough to resolve the puzzle absent HTM agents, adding HTM agents with less cyclically sensitive income (χ<1) can “re-awaken the FGP,” while adding HTM agents with more cyclically sensitive income (χ>1) can push a borderline case toward resolving it (Section 6.2, p. 26). The reason is that χ shifts the effective discounting parameter Θ̃ toward or away from 1 without changing whether the underlying risk process is procyclical, so it can tip an economy across the boundary that governs whether announcements further in the future are more or less effective than near-term ones.

It is the term -Λμ̂_{t+1} in the linearized aggregate Euler equation (Section 4, eq. 20-21): tighter monetary policy today raises expected future values of μ (the common marginal propensity to consume, which rises when real rates are low because consumption becomes more responsive to income), and a higher μ means a given level of income risk translates into more consumption risk, strengthening precautionary saving and reducing current demand. The authors call this “a novel channel of monetary policy: tighter monetary policy increases the sensitivity of individual consumption to individual income shocks, raising consumption risk and reducing demand via the precautionary savings channel… in addition to the standard intertemporal substitution channel” (Section 4, p. 15) – and note it is a direct consequence of allowing strictly positive liquidity, since it would be absent in the zero-liquidity limit used by much of the tractable-HANK literature.

The paper shows fiscal multipliers in a multi-period liquidity trap are larger under countercyclical risk and smaller under procyclical risk, mirroring the pattern found for the forward-guidance puzzle (Section 6.3, Figure 4, pp. 28-29): “procyclical income risk makes indeterminacy less likely, weakens the forward guidance puzzle and reduces the government spending multiplier in a liquidity trap. Countercyclical risk makes indeterminacy more likely, worsens the forward guidance puzzle and increases the fiscal multiplier in a liquidity trap” (Section 7, p. 29). MPC heterogeneity again interacts multiplicatively: introducing HTM agents amplifies the deviation from RANK when χ>1 and dampens it when χ<1, but “only when χ>1” does the deviation actually amplify relative to the no-HTM benchmark (Section 6.3, p. 28).

Beyond the well-known active/passive fiscal-monetary regime question of Leeper (1991) and the fiscal theory of the price level – which the paper deliberately sets aside by assuming fiscal policy always adjusts to ensure solvency – the authors identify a distinct channel: “by affecting both the cyclicality of income risk and cyclical sensitivity of the income of constrained agents, fiscal policy determines the way in which a HANK economy deviates from the corresponding RANK benchmark” (Section 7, p. 29). Ordinary tax-and-transfer design (how progressively taxes and transfers vary with the cycle) is therefore not a side issue but a first-order input into whether monetary policy in a HANK economy behaves like RANK, amplifies RANK’s predictions, or reverses them – a “Θχ” monetary-fiscal interaction the authors distinguish explicitly from both the Leeper-style solvency question and the Ricardian-equivalence-breaking channel emphasized by Kaplan, Moll and Violante (2016) (Section 7, p. 30, fn. 38-39).