An adaptive estimator is an efficient estimator for a model that is only partially specified.

We study the possible impact of daylight-saving time adjustment on stock returns. Previous work reveals that average returns tend to decline following an adjustment. As averages are sensitive to outliers, more recent work focused on the entire distribution of returns and found little impact following adjustments. Unfortunately, the general nature of the alternative hypothesis reduces the power of the distribution test to detect an effect of adjustments on the location of the distribution. We construct robust tests that are designed to have power to detect a time-adjustment effect on the location of returns. We also develop a more novel test of exponential tilting that is designed to accommodate possible heterogeneity in the return distribution over time. When we apply these test to S&P 500 stock returns, we are unable to rigorously detect a time adjustment effect on stock returns.

Microstructure noise contaminates high-frequency estimates of asset price volatility. Recent work has determined a preferred sampling frequency under the assumption that the properties of noise are constant. Given the sampling frequency, the high-frequency observations are given equal weight. While convenient, constant weights are not necessarily efficient. We use the Kalman filter to derive more efficient weights, for any given sampling frequency. We demonstrate the efficacy of the procedure through an extensive simulation exercise, showing that our filter compares favorably to more traditional methods.

We study the problem of obtaining accurately sized test statistics in finite samples for linear regression models where the error dependence is of unknown form. With an unknown dependence structure there is traditionally a trade-off between the maximum lag over which the correlation is estimated (the bandwidth) and the decision to introduce conditional heteroskedasticity. In consequence, the correlation at far lags is generally omitted and the resultant inflation of the empirical size of test statistics has long been recognized. To allow for correlation at far lags we study test statistics constructed under the possibly misspecified assumption of conditional homoskedasticity. To improve the accuracy of the test statistics, we employ the second-order asymptotic refinement in Rothenberg (1988) to determine critical values. We find substantial size improvements resulting from the second-order theory across a wide range of specifications, including substantial conditional heteroskedasticity. We also find that the size gains result in only moderate increases in the length of the associated confidence interval, which yields an increase in size-adjusted power. Finally, we note that the proposed test statistics do not require that the researcher specify the bandwidth or the kernel.

This dissertation consists of three works which consider estimation of economic

variables when a spatial component exists. Each essay utilizes different techniques

and methodology for working with data that can be grouped into spatial clusters.

In the first essay, I estimate the impact of air pollution events caused by wildfire

smoke on respiratory and circulatory health outcomes. Utilizing a combination

of California health data and NOAA wildfire smoke data I can estimate the impact

of exposure to wildfire smoke on health outcomes for all individuals in California.

Using inpatient data I am able to construct a measure of exposure to wildfire smoke

prior to the hospital visit, this allows for the identification of the impact of wildfire

smoke exposure on different health outcomes. I find that an additional day of smoke

exposure in a month leads to on average 11.38 additional hospital admissions for respiratory

diagnoses and an additional 3 hospital admissions for circulatory diagnoses.

This translates to an annual cost of wildfire smoke exposure in California due to respiratory

and circulatory hospital admissions of $192,316,498.

The second essay, joint with Travis Cyronek, asks the question: How does the

sharing economy affect traditional lodging markets? The advent of platforms such

as Airbnb in 2008 has introduced a new channel of market interaction between those

with space and those who seek it. This allows for transactions of lodging services

that might otherwise be underutilized. This paper develops a framework to help

think about how peer-to-peer transactions interact with traditional rental markets,and what this means for property managers and tenants. Specifically, we examine

how the introduction of sharing platforms (e.g. Airbnb) affect the listing decisions of

vacant property managers and the lodging choices of dwelling seekers. The model

features landlords who choose where to list vacant properties and renters who search

for lodging. Renters can be either short or long-term, referencing how long they wish

to occupy the property. Sharing platforms give landlords the option of accessing these

short-term renters whom would otherwise occupy hotels, affecting traditional, longterm

renters. We find that Airbnbs decrease hotel prices by about $24 while they

increase average rents by $39 per month.

In the third essay, joint with Douglas G. Steigerwald, we study the behavior of

cluster-robust test statistics in models with instrumental variables when cluster heterogeneity

is present. Inference in a large number of papers using two-stage least

squares regressions published in American Economics Association journals are driven

by the presence of one or two influential clusters. We link a measure of cluster heterogeneity,

the feasible effective number of clusters, to measures of influence. Using

simulations, we demonstrate that high levels of cluster heterogeneity lead to coverage

of less than 95% for 95% confidence intervals when using instrumental variables with

panel data or with data that can be grouped into clusters. Using data from papers with

two-stage least squares regressions published in American Economic Association journals,

we show that the feasible effective number of clusters can be used as a pre-test

to the sensitivity of two-stage least squares inference to influential clusters. We further

show that when the feasible effective number of clusters is small, even when the

number of clusters is large, the distribution of the test statistic in non-normal. When

this severe cluster heterogeneity is present, the restricted wild cluster bootstrap can

be used to return coverage to the appropriate level.