Introduction

Most American taxpayers cannot accurately state their marginal tax rate. Gideon (2017) finds that only 33.7% of respondents correctly understand that their marginal rate exceeds their average rate---a basic structural feature of a progressive tax system. Rees-Jones & Taubinsky (2020) document that taxpayers’ perceived marginal rates deviate from their true rates with a standard deviation on the order of 12 percentage points, and that these errors are not confined to the financially unsophisticated. The source of this confusion is not ignorance of tax law in the abstract but rather the complexity of the code itself: federal brackets interact with state income taxes, payroll tax phase-outs, the earned income tax credit (EITC), child tax credit phase-ins and phase-outs, and the alternative minimum tax (AMT) to produce effective marginal rates that no taxpayer can compute in their head.

This paper asks a simple question: how much does this misperception cost? If workers choose their labor supply based on a noisy signal of their true marginal rate, they will systematically work too much or too little relative to the optimum. The resulting utility loss---aggregated across the workforce---constitutes a deadweight loss from tax complexity that is distinct from, and additional to, compliance costs and the standard Harberger triangles of distortionary taxation.

I develop a model in which a worker has quasilinear-isoelastic preferences:

where $C$ is consumption, $h$ is hours of labor, $\varepsilon$ is the Frisch elasticity of labor supply, and $\psi$ is a scale parameter. Given a linear tax rate $\tau$ and hourly wage $w$, the budget constraint is $C = w(1-\tau)h$, and optimal hours are $h^{*} = \bigl(w(1-\tau)/\psi\bigr)^{\varepsilon}$. Now suppose the worker does not observe $\tau$ but instead receives a signal $\hat{\tau} = \tau + \delta$, where $\delta \sim N(0, \sigma^{2})$. The worker optimizes against $\hat{\tau}$ but realizes utility at the true rate. The individual deadweight loss is $U(h^{*}) - U(\hat{h})$, where $\hat{h}$ denotes hours chosen under the perceived rate. A second-order Taylor expansion yields the expected per-worker deadweight loss:

This formula has an intuitive structure. Deadweight loss is proportional to the labor supply elasticity $\varepsilon$, because more elastic workers distort their behavior more in response to misperceived prices. It is proportional to $\sigma^{2}$, reflecting the variance of the misperception error---the noise in the signal. And it is inversely proportional to $1-\tau$, the net-of-tax share, because a given absolute misperception error represents a larger proportional distortion when the net-of-tax wage is smaller.

I calibrate the model using three empirical inputs. First, the Frisch elasticity of labor supply is set to $\varepsilon = 0.33$, the central estimate from the meta-analysis of Chetty (2012). Second, the misperception standard deviation is set to $\sigma = 0.12$, consistent with the distribution of errors documented in Rees-Jones & Taubinsky (2020). Third, the mean marginal tax rate is $\bar{\tau} = 0.30$, drawing on Congressional Budget Office (2016) estimates of effective marginal rates that account for federal income taxes, payroll taxes, and means-tested transfers. Together with mean annual earnings of $55,000 and 160 million workers, the central estimate of aggregate deadweight loss is $30 billion per year, or 0.11% of GDP. Sensitivity analysis across a 3 $\times$ 3 grid of elasticities ($\varepsilon \in \{0.25, 0.33, 0.50\}$) and misperception levels ($\sigma \in \{0.08, 0.12, 0.15\}$) generates a range of 0.04% to 0.25% of GDP, or $10 to $71 billion.

These magnitudes represent 4--25 percent of the conventional DWL of the income tax estimated by Feldstein (1999), who places the deadweight loss of the income tax at roughly 1% of GDP. Skinner (1988) showed that uncertainty about future tax policy imposes welfare costs of similar order; the present paper complements that finding by showing that uncertainty about current rates---driven by code complexity rather than policy instability---is also costly.

This paper makes three contributions. First, it derives a closed-form expression for the expected deadweight loss from tax misperception under standard preferences, making the welfare cost transparent and easy to calibrate. Second, it provides the first calibration of this cost using empirically grounded estimates of misperception dispersion from Rees-Jones & Taubinsky (2020) and Gideon (2017), combined with standard labor supply parameters. Third, it shows that a utilitarian social planner who accounts for misperception chooses a lower optimal linear tax rate---42.9% versus 44.5% under perfect information---because the marginal welfare cost of taxation is amplified by the noise in taxpayers’ perception of rates.

The remainder of the paper is organized as follows. The model chapter formalizes the preference structure, derives the deadweight loss formula, and characterizes optimal taxation under misperception. The calibration chapter describes the empirical inputs and sensitivity analysis. The results chapter presents the baseline estimates, distributional implications, and the planner’s problem. The conclusion discusses policy implications---in particular, why reducing $\sigma$ through simplification dominates reducing $\tau$ through rate cuts---and outlines extensions to non-linear schedules and heterogeneous misperception.