Maximum entropy regressions

Maximum entropy (ME) regression is compared to ordinary regression in the case of two observations on two normally distributed variables (one dependent and one explanatory) with correlation coefficient ρ. ME regressions have the smaller risk under quadratic loss if ? lies in the interval ±0.95. In the case of two explanatory variables and two observations, ordinary regression is not possible but ME regressions do exist and have finite risk.     

The symmetric maximum entropy distribution

The symmetric maximum entropy (ME) distribution is introduced based on symmetrized order statistics. Its properties (including higher-order moments) are compared with those of the non-symmetric ME distribution.     

Structural oil price shocks and policy uncertainty

Increases in the real price of oil not explained by changes in global oil production or by global real demand for commodities are associated with significant increases in economic policy uncertainty and its four components (the volume of newspaper coverage of policy uncertainty, CPI forecast interquartile range, tax legislation expiration, and federal expenditures forecast interquartile range). Oil-market specific demand shocks account for 31% of conditional variation in economic policy uncertainty and 22.9% of conditional variation in CPI forecast interquartile range after 24 months. Positive oil shocks due to global real aggregate demand for commodities significantly reduce economic policy uncertainty. Structural oil price shocks appear to have long-term consequences for economic policy uncertainty, and to the extent that the latter has impact on real activity the policy connection provides an additional channel by which oil price shocks have influence on the economy. As a robustness check, structural oil price shocks are significantly associated with economic policy uncertainty in Europe and energy-exporting Canada.     

Further evidence on LIML/ME estimates

A simulation experiment suggests that LIML based on maximum entropy (ME) moments performs well in the case of correlated exogenous variables even when the sample is undersized. The asymptotic standard errors of LIML/ME are evaluated for the first time and appear to be accurate even in moderate sized samples.     

The positive maximum entropy distribution

The maximum entropy (ME) principle is applied to samples from a positive population, yielding a positive ME distribution with one truncated exponential component distribution. This approach is easily extended to double-constrained variables such as proportions (e.g., market shares and budget shares).     

Uncertainty and Investment Dynamics

This paper shows that with (partial) irreversibility higher uncertainty reduces the responsiveness of investment to demand shocks. Uncertainty increases real option values making firms more cautious when investing or disinvesting. This is confirmed both numerically for a model with a rich mix of adjustment costs, time-varying uncertainty, and aggregation over investment decisions and time and also empirically for a panel of manufacturing firms. These "cautionary effects" of uncertainty are large—going from the lower quartile to the upper quartile of the uncertainty distribution typically halves the first year investment response to demand shocks. This implies the responsiveness of firms to any given policy stimulus may be much weaker in periods of high uncertainty, such as after the 1973 oil crisis and September 11, 2001.     

A new test for deficit sustainability and its application to US data

In this paper, we define deficit sustainability by requiring formally that both the discounted debt vanish asymptotically and the undiscounted debt be bounded. Thus, a new necessary condition and a new testing procedure emerge. We propose a new test statistic and prove that its limiting distribution is standard normal, N(0, 1). Its finite- sample distribution differs from N(0, 1), however, mainly because it has fat tails, so we derive empirical critical values using simulations. Using the new test and United States (US) quarterly data, the conclusions of three earlier papers that fail to reject the sustainability of the US budget or current-account deficit are reversed.     

Variations of Delphi

This paper describes a number of variations to refine the classical Delphi technique, which have been tested in two recent Delphis. A calculated median and interquartile range are introduced. The use of parallel questionnaires to explore many questions simultaneously is described. The use of probabilities in ratio form is suggested, as are general data transforms for median and interquartile range calculation. The idea of ranking the timing of events is described, and when calculations of the different timescales people use are employed, it is shown that part of the convergence in succeeding rounds arises from changes in people's timescales. Correlation between events is used to identify events that interact and hence, Cross Impact Analysis is required.     

Do the world’s largest cities follow Zipf’s and Gibrat’s laws?

We examine whether the size distribution and the growth process of the world’s largest cities follow Zipf’s law and Gibrat’s law. The parametric results of the size distribution analysis reject Zipf’s law for all sample sizes and also show the Zipf exponent systematically declines as the sample size increases. The growth process analysis confirms Gibrat’s law and yields a local Zipf exponent of one for cities with a normalized population less than 0.53%, which includes about 95% of the total observations. The deviations from Zipf’s law occur at the extreme upper tail and are likely a result of restricted mobility of population across countries. However, given that Gibrat’s law holds, we can expect the size distribution to converge to Zipf’s law with a decline in the barriers to immigration.     

On estimating long-run effects in models with lagged dependent variables

A common procedure in economics is to estimate long-run effects from models with lagged dependent variables. For example, macro panel studies frequently are concerned with estimating the long-run impacts of fiscal policy, international aid, or foreign investment.Our analysis points out the hazards of this practice. We use Monte Carlo experiments to demonstrate that estimating long-run impacts from dynamic models produces unreliable results.Biases can be substantial, sample ranges very wide, and hypothesis tests can be rendered useless in realistic data environments. There are three reasons for this poor performance. First, OLS estimates of the coefficient of a lagged dependent variable are downwardly biased in finite samples. Second, small biases in the estimate of the lagged, dependent variable coefficient are magnified in the calculation of long-run effects. And third, and perhaps most importantly, the statistical distribution associated with estimates of the LRP is complicated, heavy-tailed, and difficult to use for hypothesis testing. While many of the underlying problems have been long-known in the literature, the continued widespread use of the associated empirical procedures suggests that researchers are unaware of the extent and severity of the estimation problems. This study aims to illustrate their practical importance for applied research.     

Testing for Granger non-causality in heterogeneous panels

This paper proposes a very simple test of Granger (1969) non-causality for heterogeneous panel data models. Our test statistic is based on the individual Wald statistics of Granger non causality averaged across the cross-section units. First, this statistic is shown to converge sequentially to a standard normal distribution. Second, the semi-asymptotic distribution of the average statistic is characterized for a fixed T sample. A standardized statistic based on an approximation of the moments of Wald statistics is hence proposed. Third, Monte Carlo experiments show that our standardized panel statistics have very good small sample properties, even in the presence of cross-sectional dependence.     

The product life cycle and sample representativity bias in price indexes

Official price indexes are usually calculated using matched samples of products. If products exhibit systematic price trends at different points in their life cycle then matched sample methods may introduce bias if the life cycle movement in the sample does not adequately reflect that in the population. This article explores the extent of these life cycle pricing effects and then examines the bias it can introduce in measured inflation. A large US supermarket scanner data set for six cities and six products over 12 years is used. Using hedonic methods we find that the life cycle component of price change is important across a range of products and cities. To explore the bias introduced by these movements, we use simulations that construct indexes with different sample update frequency. For indexes that are never completely resampled, we find an annual absolute bias of 0.88 and 0.59 percentage points depending upon whether we use the actual prices or prices imputed from our hedonic model. This compares with absolute biases of 0.34 and 0.10 percentage points for the corresponding cases for samples, which are re-selected annually. Thus our results provide strong support for more frequently updating index samples.     

Economic models based on Pareto and Gamma random variables

The distribution of products of random variables arises explicitly in economics and related areas. This has increased the need to have available the widest possible range of statistical results on products of random variables. In this note, the exact distribution of the product XY is derived when X and Y are independent random variables and arise from the two most applied models for economic data. The associated estimation procedures and percentage points are provided.     

Understanding alcohol consumption across countries

ABSTRACT

With unpublished data from the International Comparison Program that cover the consumption of three alcoholic beverages in over 150 countries, we analyse drinking patterns around the world with an index-number approach, by estimating a demand system, and by studying the interaction among beverages in generating utility. We consider a separate demand system for each income quartile and find that tastes are not too different across quartiles. Broadly speaking, the results are robust to rolling sub-samples of countries, an alternative demand model and sample selectivity issues. The differences in the cost of alcohol across countries are also investigated, as is its role in affecting the degree of price-sensitivity of consumption.     

Do insurance activities enhance the performance of financial services holding companies?

Stiroh and Rumble (2006) and Yeager et al. (2007) have argued that extension of banks into nonbanking activities produces no diversification benefits for financial services holding companies (FSHCs) eligible to consolidate banking and insurance products. We investigate the effect of aggregate nonbank activities, as well as two main areas of insurance business, namely agency and underwriting enterprise, on the risk-adjusted performance of FSHCs. Our sample includes a quarterly panel data of FSHCs over the period 2003–2011. We find that the effect of FSHCs’ entry into agency insurance activities on their risk-adjusted returns is dependent on their size; it is positive for FSHCs in the top quartile of our sample (assets greater than $4.495 billions), but negative for smaller and mid-size FSHCs. The effect of underwriting insurance activities on risk-adjusted returns is generally negative or statistically insignificant, except for some of the largest-size firms.     

Growth,Death, and Taxes

Recent theories of endogenous growth suggest that changes in tax rates may permanently affect growth. However, attempts to quantify these growth effects have reached very different conclusions in spite of a common theoretical framework: the neoclassical growth model with human capital accumulation by infinitely lived households. This paper shows that a model which explicitly specifies human capital accumulation over the life-cycle provides sharper answers. In such a model, a plausible range for the growth effects of eliminating taxes in the United States is between 0.5 and 1.3 percentage points compared with 0 to 4 percentage points in the infinite horizon model. The much wider range found in the literature is due to two assumptions which are commonly viewed as innocuous simplifications but contrast sharply with traditional human capital theory: that households are infinitely lived and face constant point-in-time returns in human capital accumulation. The widely held view that long, finite horizons are closely approximated by infinite horizons is generally invalid. Abstracting from finite horizons leads to a systematic overstatement of the growth effects of taxes. Journal of Economic Literature Classification Numbers: J24, O41.     

An augmented autoregressive distributed lag bounds test for cointegration

An augmented autoregressive distributed lag (ARDL) bounds test for cointegration involves an extra F-test on the lagged levels of the independent variable(s) in the ARDL equation. Originally, this testing strategy was introduced using the bootstrap procedure. This paper provides both the small sample and asymptotic critical values for easier implementation of the test, making it applicable for a broader range of researchers. Two advantages of this augmented ARDL bounds test are that the assumption of an I(1) dependent variable is not necessary, and a clear conclusion on the cointegration status is provided by the three tests. The augmented ARDL bounds test is demonstrated using an empirical study on government taxation and expenditures. The tests support the tax-and-spend hypothesis of the budgetary policy for the US, the UK, and France.     

Changes in the wage structure,family income,and children's education

We exploit the changes in the distribution of family income to estimate the effect of parental resources on college education. Our strategy exploits the fact that families at the bottom of the income distribution were much poorer in the 1990s than they were in the 1970s, while the opposite is true for families in the top quartile of the distribution.  Our estimates suggest large effects of family income on enrollments.  For example, we find that a 10 percent increase in family income is associated with a 1.4 percent increase in the probability of attending a four-year college.     

Semiparametric and nonparametric testing for long memory: A Monte Carlo study

The finite sample properties of three semiparametric estimators, several versions of the modified rescaled range, MRR, and three versions of the GHURST estimator are investigated. Their power and size for testing for long memory under short-run effects, joint short and long-run effects, heteroscedasticity andt-distributions are given using Monte Carlo methods. The MRR with the Bartlett window is generally robust with the disadvantage of a relatively small power. The trimmed Whittle likelihood has high power in general and is robust except for large short-run effects. The tests are applied to changes in exchange rate series (daily data) of 6 major countries. The hypothesis of no fractional integration is rejected for none of the series.     

Migrants as second-class workers in urban China? A decomposition analysis

In urban China, urban resident annual earnings are 1.3 times larger than long-term rural migrant earnings as observed in a nationally representative sample in 2002. Using microsimulation, we decompose this difference into four sources, with particular attention to path-dependence and statistical distribution of the estimated effects: (1) different allocation to sectors that pay different wages (sectoral effect); (2) hourly wage disparities across the two populations within sectors (wage effect); (3) different working times within sectors (working time effect); and (4) different population structures (population effect). Although sector allocation is extremely contrasted, with very few migrants in the public sector and very few urban residents working as self-employed, this has no clear impact on earnings differentials, because the sectoral effect is not robust to the path followed for the decomposition. The second main finding is that the population effect is robust and significantly more important than wage or working time effects. This implies that the main source of disparity between the two populations is pre-market (education opportunities) rather than on-market.