**6. Findings**

Tables 2 and 3 present the results. Notice in Table 2, the three different estimator methods give almost the same result. With the sample used in this analysis, the choice of estimator method has little impact on the conclusion in the testing of Gibrat's LPE. Consequently, the focus will be on the estimates from the SYS-GMM in the subsequent discussion, as this is the most widely used estimator for these purposes.

**Table 2.** Dynamic panel data estimators testing Gibrat's Law for Norwegian campsites over a 10-year period (2010–2019) (Robust Standard error in parentheses).


Notes: AR(1) tests H1 (β = 1), MA(1): H2 (ρ = 0) and Heteroscedasticity(t): H3 (δ = 0). Not possible to estimate AR(2) for large firms \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1.


Notes: Mod 1 is without control variables (see Table 1). Mod 2 includes control variables. \*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1.

Hypothesis H1 is rejected in favour of β < 1 for the whole sample with a significance level of 5%, and for medium campsites with a significance level of 1%. For large firms, we cannot reject the hypothesis β = 1; there is no evidence that they do not follow a random walk. The coefficient is close, but below unity for this subsample. We cannot reject a random walk for the smallest firms either, but this is not due to the coefficient being close to unity. Instead, the standard error is too large. This indicates that the smallest campsites have large variations in their growth paths, where most revert quickly to their mean, whereas others may follow random walks, and still others may even have explosive growth paths. The same dynamic of small, medium, and large campsites, where there seems to be a threshold size where the campsites' path changes, is also reflected in the moving average parameter, the heteroskedasticity tests, and the autocorrelation tests.

Firstly, from the size-related heteroskedasticity test, we can see that, for the whole sample—and the small and medium firms—the variation in their size (revenue) decreases the larger the firms are. This indicates that there may be a threshold size for campsites, where additional size does not translate into more stable revenue.

We see the same threshold dynamic in the moving average component, where the only significant MA component is found in the large firms. The revenue-deviations of large campsites spill over from one year to another, one year's success being a significant predictor of success in the following year. The success or failure of large campsites in one year persists into the next year, whereas the success/failure of small and medium campsites is absorbed into their revenue in the year the success/failure happens. That is, shocks that occur to large campsites have a certain inertia as it pertains to their size (revenue). This is reflected in the test for second year autocorrelation, which is only significant for large campsites. Consequently, the moving average parameter, the heteroskedasticity tests and the autocorrelation test all point to there being a threshold size for campsites at which their growth paths change.

The absolute average value dependence across the sample is 34.5%, and it is quite stable regardless of the size of the campsites. One reason might be competition between campsites, whereas another is that they are all affected by the same market forces and consumer tastes. Model 2 includes two other independent variables (exchange rate and debt; see Table 2). Both have the expected positive signs, and both are significant at the one percent level. A depreciation in the exchange rate has been shown to lead to more foreign visitors, but it might also lead to more Norwegians choosing to stay in the country for their vacation.
