A Replication Study of ‘Why Do Cities Hoard Cash?’ (The Accounting Review, 2009) ABSTRACT Gore’s article explores the determinants and implications of cash reserves. We first attempted to replicate Gore’s finding of a positive relationship between environmental uncertainty and municipal fund balances (2009) using the same data, the same specifications, and the same econometric software. We then tested the robustness of her original findings by adding years and observations. We show that the empirical results reported in this article are largely replicable and that its results are robust to substantial data extensions. Nevertheless, we believe that Gore reaches normative conclusions, that municipalities hold “excess cash reserves,” which are not justified by her empirical results. Keywords: Reserves • Volatility • Replication JEL Classification Numbers: H71 • H72
18
Embed
A Replication Study of ‘Why Do Cities Hoard Cash?’ (The Accounting Review, 2009)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A Replication Study of ‘Why Do Cities Hoard Cash?’ (The Accounting Review, 2009)
ABSTRACT
Gore’s article explores the determinants and implications of cash reserves. We first attempted to
replicate Gore’s finding of a positive relationship between environmental uncertainty and
municipal fund balances (2009) using the same data, the same specifications, and the same
econometric software. We then tested the robustness of her original findings by adding years
and observations. We show that the empirical results reported in this article are largely
replicable and that its results are robust to substantial data extensions. Nevertheless, we believe
that Gore reaches normative conclusions, that municipalities hold “excess cash reserves,” which
are not justified by her empirical results.
Keywords: Reserves • Volatility • Replication
JEL Classification Numbers: H71 • H72
1
A Replication Study of ‘Why Do Cities Hoard Cash?’ (The Accounting Review, 2009)
1. INTRODUCTION
The Government Finance Officer’s Association recommends that municipalities maintain
reserves at least equal to about 16 percent of revenues, plus more to deal with revenue
volatility, infrastructure upkeep and vulnerability to extreme events. Kriz (2002) and Dothan
and Thompson (2009) argue that they should (as a normative matter) increase reserves (fund
balances) in line with revenue volatility. Indeed, Kriz concluded that if the representative
Minnesota municipality wished “to sustain a three percent expenditure growth rate with a 75
percent confidence level, it would need savings equal to 91 percent of total revenues” (Kriz
2002: 5).
Angela Gore’s 2009 article in Accounting Review is especially important because it shows
that local-government fund balances do apparently vary directly with revenue volatility and
that jurisdictions that spend more on administration tend to maintain higher reserves. These
finding are critical to the developing field of public financial management. Consequently, we
wished to pursue them further, especially since we had reservations about Gore’s data set,
specification of response and predictor variables, and functional forms tested. Unfortunately,
her data set and codes were unavailable. Consequently, we set out to replicate her work, as a
first step as precisely as possible, using the same data, the same specifications, and the same
statistical software1 (Stata). Next, we extended the time horizon of her analysis to include all of
the years of data available.
1Gore used SAS to organize (collate and clean) her data and Stata to analyze it.
2
We also briefly address her article’s fundamental hypothesis: that municipalities over
save, i.e., hold more cash than is needed to “provide a constant level of services to citizens,
regardless of revenue volatility” (Gore 2009, 183).
2. REPLICATION OF SAMPLE SELECTION AND DATA CLEANING
Starting from the government finance database2 (Pierson et al. 2014), which has data from the
Census’s annual survey of state and local governments for years between 1967 and 2011, we
restricted our sample to governments with data from years between 1997 and 2003.
Gore does not explicitly identify the government type codes that she includes in her data
set, but it appears that her analysis comprehends both municipalities (type 2) and townships
(type 3). Table 1 shows the breakdown of the data by year and type of government. It is clear
from this table that using only one government type is too restrictive.
Table 1: Goes about here
Including both municipalities and townships allows us to come close to Gore’s count of
80,125 observations. Unfortunately there is no reasonable way to replicate this number
precisely. Gore may have been working from Census data that had yet to be finalized since the
more recent data from the census includes additional data points.
Gore next drops “4,043 observations with missing data for cash or operating expenses, and
57 observations with apparent errors such as negative debt.” We adopt Gore’s definition of cash
and securities and drop 6,547 observations that have missing values for this variable. We also
drop 505 observations with missing data for total operating expenditure.
It is unclear how Gore calculates total debt from the census data, especially considering
the fact that none of the top-level debt outstanding line items in our data have negative entries.
Table 3: Winsorized Sample Standard Deviations Compared to Gore’s (2009) Table 3 Gore Replication Percent Difference Variable Full Small Full Small Full Small
Table 5: Regression Results Following Gore’s Table 4 Model 1 Gore Replication Same
Sign Variable Slope T Slope t
Intercept 19.95 10.01 30.80 13.33 Yes CV Revenue 7.92 6.00 7.39 4.38 Yes Debt per Capita t-1 -0.24 -2.72 -0.05 -0.39 - Limited Revenue 21.74 8.91 - - - Size -0.94 -10.07 -1.44 -10.87 Yes Growth 12.46 7.00 4.20 2.90 Yes State Revenue -3.89 -3.06 -8.77 -5.26 Yes Quarter dummies Included Included Year dummies Included Included State dummies Included Included Adj. R2 0.21 0.19 Sample Size 9,413 9,576 Note: Results of a replicated regression modeling months of cash reserves according to equation 1. The standard
errors used to calculate t-statistics for both Gore's regressions and our replication are robust and clustered by
government. The slopes we show in bold are significant at the 5 percent level or better.
15
Table 6: Regression Results Following Gore’s Table 5 Model 1 Gore Replication Same Sign Variable Slope T Slope T
Intercept 0.44 19.97 0.50 22.15 Yes Excess Cash t-1 0.01 7.87 0.001 4.30 Yes Debt per Capita -0.01 -5.36 0.0003 0.24 - Size -0.02 -14.24 -0.02 -11.11 Yes Year dummies Included Included State dummies Included Included Adj. R2 0.25 0.22 Sample Size 7,379 4,791 Note: Results of a replicated regression modeling months of cash reserves according to equation 2. The standard
errors used to calculate t-statistics for both Gore's regressions and our replication are robust and clustered by
government. The slopes we show in bold are significant at the 5 percent level or better.
16
Table 7: Regression Results Following Gore’s Table 4 Model 1 Using All of the Data Gore Replication Same Sign Variable Slope T Slope t
Intercept 19.95 10.01 26.97 Yes CV Revenue 7.92 6.00 1.27 5.46 Yes Debt per Capita t-1 -0.24 -2.72 0.189* 1.93 Sign Change Limited Revenue 21.74 8.91 - Size -0.94 -10.07 -2.24 -47.83 Yes Growth 12.46 7.00 1.85 9.71 Yes State Revenue -3.89 -3.06 -4.03 -10.03 Yes Quarter dummies Included Included Year dummies Included Included State dummies Included Included Adj. R2 0.21 0.13 Sample Size 9,413 389,365 Note: Results of a replicated regression modeling months of cash reserves according to equation 1, but including
all of the available data. The standard errors used to calculate t-statistics for both Gore's regressions and our
replication are robust and clustered by government. The slopes we show in bold are significant at the 5 percent
level or better. A * signifies significance at the 10 percent level, but not the 5 percent level.
17
Table 8: Regression Results Following Gore’s Table 5 Model 1 Using All of the Data Gore Replication Same Sign Variable Slope T Slope t
Intercept 0.44 19.97 0.77 107.12 Yes Excess Cash t-1 0.01 7.87 0.0012 30.88 Yes Debt per Capita -0.01 -5.36 -0.005 -6.52 - Size -0.02 -14.24 -0.04 -72.40 Yes Year dummies Included Included State dummies Included Included Adj. R2 0.25 0.22 Sample Size 7,379 387,222 Note: Results of a replicated regression modeling months of cash reserves according to equation 2, but including
all of the available data. The standard errors used to calculate t-statistics for both Gore's regressions and our
replication are robust and clustered by government. The slopes we show in bold are significant at the 5 percent