The Effect on Growth of a Tax Shift between Land Value Taxes and Income Taxes in Denmark

Show more

1. Introduction

The New Physiocratic League advocates for a platform to shift taxation away from personal & corporate income taxation, and primarily towards a particular form of land value tax (LVT). It also advocates for anti-corruption measures and monetary reforms among other proposals. This study focuses specifically on the results of a shift to an LVT.

Empirical research on true Georgist policy, namely a tax system where the tax burden falls mainly on land, is limited due to a lack of real-world examples. The examples that have existed, mainly lack data. However, Denmark has an explicitly Georgist party which had success in parliament in the 1950s and helped implement a high national rate of land value tax where much of the tax burden existed at the time. Denmark provides a rare example where one can assess economic performance from this Georgist period, gradually transitioning to today where the tax burden falls mainly on incomes. While ample theoretical research exists on the implementation of an LVT, this study is the first of its kind to look at empirical data.

The remainder of this paper is organized as follows: In Section 2 the data and methods used in this study are presented. Section 3 includes stationary testing of time series. The estimated model and its checking are presented in Section 4 and Section 5, respectively. Concluding remarks are offered in the final section.

2. Data Source and Methods

All data for this study is taken from Statistics Denmark (the central authority on Danish statistics), and spans from 1966 to 2018. This brings to light another advantage of using LVT data from Denmark versus other countries using LVT, in that they have a fully complete set of tax and economic performance data.

The yearly data on gross domestic product (gdp), household consumption expenditure (housc), government consumption expenditure (gov), total taxes and duties (total_tax), total income taxes (income_tax), and land value taxes (lvt) in Denmark from 1966-2018 have been obtained^{1}.

For analysis purposes, the following variables have been derived from the data: gdp growth rate (gdp_gr), household consumption expenditure growth rate (c_gr), government consumption expenditure growth rate (gov_gr), the proportion of income tax in total taxes (inc_div_total), and the proportion of LVT in total taxes (lvt_div_total)^{2}. Below are the formulas:

$gdp\_g{r}_{t}=\frac{gd{p}_{t}-gd{p}_{t-1}}{gd{p}_{t-1}}$ (1)

$c\_g{r}_{t}=\frac{hous{c}_{t}-hous{c}_{t-1}}{hous{c}_{t-1}}$ (2)

$gov\_g{r}_{t}=\frac{go{v}_{t}-go{v}_{t-1}}{go{v}_{t-1}}$ (3)

$inc\_div\_tota{l}_{t}=\frac{income\_ta{x}_{t}}{total\_ta{x}_{t}}$ (4)

$lvt\_div\_tota{l}_{t}=\frac{lv{t}_{t}}{total\_ta{x}_{t}}$ (5)

where $t$ stands for time period.

The summary statistics of the variables are presented in Table 1.

Furthermore, the time series under study have been tested for stationarity

Table 1. Descriptive statistics of variables.

applying Augmented Dickey-Fuller (ADF) test, and an autoregressive distributed lag (ARDL) model has been estimated, with Eviews as the tool of choice.

3. Testing for Stationarity

To test the time series for stationarity ADF test has been applied. Otherwise, a time series that is nonstationary can only be studied on its behaviour only for a specific episode. Consequently, it will not be possible to generalize it for other time periods or making predictions^{3}. The results are shown in Table 2.

The null hypothesis states that the variable under study has a unit root. As we can see, even at 10% significance level neither time series is stationary, except for gdp_gr and gov_gr. Therefore, the first order differences of nonstationary variables (d(c_gr), d(inc_div_total), d(lvt_div_total)) have been tested for unit roots. According to the results, the first order differences of all the nonstationary time series under study are stationary even at 1% significance level.

4. Model Estimation

To begin with, the best ARDL model based on Akaike information criterion (AIC) has been selected using automatic selection option in Eviews. ARDL (1, 0, 0, 4, 0) model fits the data best according to AIC. The results are shown in Table 3.

According to the estimation results, some parameters are not statistically significant even at 10% level. Furthermore, the non-significant variables have been eliminated from the model. The estimation results of the new model are shown in Table 4.

As shown, now all parameters are statistically significant even at 1% significance level (except for the constant). Besides, Akaike information criterion has

Table 2. Augmented Dickey-Fuller (ADF) test results for the time series under study^{4}.

Table 3. Estimation output.

been improved ( $AIC=-5.58$ ). For comparison purposes, the AIC for the previous model was −5.45

Table 4. Estimation output.

Thus, the estimated equation is:

$\begin{array}{l}GDP\_GR=0.36\ast GDP\_GR\left(-1\right)+0.23\ast GOV\_GR+0.22\ast GOV\_GR\left(-4\right)\\ \text{}-4.8\ast D\left(LVT\_DIV\_TOTAL\right)+0.42\ast D\left(C\_GR\right)+0.007\end{array}$

That is, the GDP growth rate in Denmark at time
$t$ , GDP_GR, is explained by its value at time
$t-1$ , by the rate of change of government expenditure at time

It is worth noticing that the difference of LVT ratio at time t has a negative impact on GDP growth rate for the same period of time.

5. Model Diagnostic Checking

The estimated residuals from the regression have been tested for autocorrelation. As we can see from Figure 1 all values of the correlogram are within the boundaries. To conclude, we fail to reject the null of no autocorrelation between the residuals. So there are no significant autocorrelations between residuals according to the ACF plot.

Furthermore, the errors normality and homoscedasticity assumptions have been checked.

The results are as follows:

As shown above in Figure 2, at 5% significance level the residuals do not appear to be normally distributed (null hypothesis: errors are normally distributed).

The output above (Table 5) suggests not rejecting the null of homoskedasticity at 5% significance level. So, we can conclude there is no heteroskedasticity.

Figure 1. Correlogram of residuals.

Figure 2. Jarque-Bera normality test for residuals.

Table 5. Test for Heteroskedasticity.

Table 6. Serial correlation LM test.

Figure 3. CUSUM test.

Figure 4. CUSUM of squares test.

Table 6 results reveal that the Breusch-Godfrey Serial Correlation LM Test^{5} fails to reject the null at 5% significance level.

Furthermore, the cumulative sum of recursive residuals (CUSUM) and the CUSUM of squared residuals (CUSUMSQ) tests have been applied in order to test for the stability of the estimated parameters.

Figure 3 clearly indicates no coefficient instability in the equation during the sample period as the cumulative sum is inside the 5% critical lines. Likewise, Figure 4 shows that the cumulative sum of squares is also within the 5% critical lines, suggesting that the residual variance is stable.

6. Conclusion

A shift towards a land value tax is the only solution to many of the growth and taxation issues governments are facing. However, it is important that tax reform is delivered in as few packages as possible, as the economy appears to respond best when tax rates are predictable, as in Denmark. The LVT-based platform offered by the New Physiocratic League, if implemented, would best ensure real progress if put into place quickly, and then allowed the economy time to reorient itself towards its new trajectory of sustainable yet stronger growth.

NOTES

^{1}source: https://www.dst.dk/en [1].

^{2}The proportion of taxes (not the absolute values) are used since it is very challenging to figure out if the change in absolute value is caused by economic activity change or some tax policy change [2].

^{3}Gujarati, Damodar N., and Dawn C Porter. Basic Econometrics. 5th ed. Boston: McGraw-Hill, 2008 [3].

^{4}**, **, and *** represent significance at the 10% (p < 0.1), 5% (p < 0.05), and 1% (p < 0.01) statistical levels, resprctively.

^{5}In this case the Durbin-Watson statistic is not applicableas we a have lagged dependent variables on the right side of the model.