We develop a test for instrument strength based on the bias of two-stage least squares (2SLS) that (1) generalizes Stock and Yogo’s and Sanderson and Windmeijer’s tests to be robust to heteroskedasticity and autocorrelation, and (2) extends Montiel Olea and Pflueger’s robust test for models with a single endogenous regressor to multiple endogenous regressors. Our test can be based either on an absolute bias criterion or on the 2SLS bias relative to a worst-case benchmark. We also develop extensions to test whether weak instruments cause bias in individual 2SLS coefficients. In simulations, our test controls size and is powerful, and we provide efficient code packages for its practical implementation. We demonstrate our testing procedures in the context of the estimation of state-dependent fiscal multipliers, following recent leading estimates.

I propose to identify an SVAR, up to shock ordering, using the autocovariance structure of the squared innovations implied by an arbitrary stochastic process for the shock variances. These higher moments are available without parametric assumptions on the variance process. In contrast, previous approaches exploiting heteroskedasticity rely on the path of innovation covariances, which can only be recovered from the data under specific parametric assumptions on the variance process. The conditions for identification are testable. I compare the identification scheme to existing approaches in simulations and provide guidance for estimation and inference. I use the methodology to estimate fiscal multipliers peaking at 0.86 for tax cuts and 0.75 for government spending. I find that tax shocks explain more variation in output at longer horizons. The empirical implications of my estimates are more consistent with theory and the narrative record than those based on some leading approaches.

I propose to identify announcement-specific decompositions of asset price changes into monetary policy shocks exploiting heteroskedasticity in intraday data, accommodating both changes in the nature of shocks and the state of the economy across announcements. I compute decompositions with respect to Fed Funds, forward guidance, asset purchase, and Fed information shocks from January 1996 to December 2019. The decompositions illustrate which announcements of unconventional policy measures had significant effects during the Great Recession. Overall, forward guidance and asset purchases have significant effects on yields, spreads, equities, and uncertainty, but the effects of monetary policy vary over time, particularly asset purchases.

Identification via heteroskedasticity exploits variance changes between regimes to identify parameters in simultaneous equations. Weak identification occurs when shock variances change very little or multiple variances change close to proportionally, making standard inference unreliable. I propose an F-test for weak identification in a common simple version of the model. More generally, I establish conditions for validity of nonconservative robust inference on subsets of the parameters, which can be used to test for weak identification. I study monetary policy shocks identified using heteroskedasticity in high-frequency data. I detect weak identification, invalidating standard inference, in daily data, while intraday data provide strong identification.

Heteroskedasticity‐ and autocorrelation‐robust (HAR) inference in time series regression typically involves kernel estimation of the long‐run variance. Conventional wisdom holds that, for a given kernel, the choice of truncation parameter trades off a test's null rejection rate and power, and that this tradeoff differs across kernels. We formalize this intuition: using higher‐order expansions, we provide a unified size‐power frontier for both kernel and weighted orthonormal series tests using nonstandard “fixed‐ b ” critical values. We also provide a frontier for the subset of these tests for which the fixed‐ b distribution is t or F . These frontiers are respectively achieved by the QS kernel and equal‐weighted periodogram. The frontiers have simple closed‐form expressions, which show that the price paid for restricting attention to tests with t and F critical values is small. The frontiers are derived for the Gaussian multivariate location model, but simulations suggest the qualitative findings extend to stochastic regressors.