The Impact of Economic Factors on Medium-Term Budget Revenue Forecasts: Insights from an Ex Post Analysis of Advanced Economies

Abstract

This study investigates the determinants of medium-term revenue forecast errors across eight developed countries: the United States, the United Kingdom, Germany, Ireland, Hong Kong, New Zealand, Australia, and Canada. By examining two- and three-year revenue forecasts, this study applies the Kenneth Holden–David Peel test to identify forecast biases and employs panel regression models to assess the economic factors influencing forecast accuracy. The findings indicate that inflation, GDP growth, and budget balance rules are positively associated with forecast errors, whereas unemployment, population growth, and the number of fiscal rules mitigate these errors. Panel estimates reveal that fiscal structure-related variables are not only statistically significant but also economically meaningful determinants of medium-term revenue forecast errors. The results underscore the persistent challenges in achieving accurate revenue forecasts and highlight the necessity for improved forecasting methodologies to enhance fiscal policy effectiveness and resource allocation. Strengthening forecasting frameworks can contribute to more reliable revenue projections, reducing fiscal uncertainty and supporting sound economic decision making.

1. Introduction

The medium-term budgeting method is a budgeting system where yearly budgets are analyzed together with multi-year revenue and expenditure forecasts. Within this scheme, budget laws cover not only annual budgets but also estimates for the next two or more years. The correctness of these estimates is of great importance, especially regarding the future efficient financing of large public expenditures (Auld, 1970).

In the present study, the major causes of revenue forecast errors at the medium-term level in eight developed nations, namely, the United States, the United Kingdom, Germany, Ireland, Hong Kong, New Zealand, Australia, and Canada, are studied. Revenue forecasts for periods of two and three years are considered independently. Initially, the Holden and Peel (1990) test was conducted to assess whether there was bias in the forecast errors. Then, two separate panel regression models were developed to identify the economic factors affecting the two- and three-year forecast errors. The independent variables in this analysis include inflation, unemployment, the population growth rate, the GDP, government expenditures, economic crisis periods, budget balance rules, the number of fiscal rules, and the current account balance.

The findings shed light on some of the more significant results in relation to the revenue forecast errors of the countries in question. While forecast bias has manifested excessively only in the United States, this affects both two-year and three-year forecasts. The panel regression analysis pointed to a positive correlation between increases in inflation and GDP and forecast errors, while the existence of budget balance regulations contributed more to discrepancies. It appears now that inflation, GDP growth, and balance rules have higher values that yield high errors in forecasting, while rising values of unemployment, population growth, and the number of fiscal rules diminish forecast errors. Unlike the former two, crisis periods, government spending, and current account balances are very insignificant in both models. Population increase has further ceased to be statistically tactful in predicting three-year forecasts.

The assessment indicates that while developing revenue forecasts is an arduous task, bias evaluation does spell an end for them in the United States. It directly asks for improved approaches, as issues hardly ever find a solution without being addressed; on the contrary, the burgeoning influences of inflation, GDP growth, and rules concerning balanced budgets call for such vigilance by policymakers that no changes can escape their attention.

In short, the negative influences of unemployment, population growth, and the number of numerical rules on forecast errors are expressive of the need for a more detailed understanding of these variables to aid in enhancing forecasting models. Statistics denote that in times of emerging crises, government expenditures and current account balances come together in their non-existing significant effects on revenue predictions. In this sense, the outcome strongly urges one to adopt a trickier approach as far as the scope is concerned, which meanwhile produces a number of economic indicators classified into different categories for revenue forecasting, thereby enabling policymakers to see them in the light of various categories. As a result, the accuracy of these forecasts turns out to be a decisive factor not just for developed countries’ good fiscal planning but also for their sustainable economic growth.

1.1. Theoretical Rationale for Medium-Term Budgeting

Public budgeting constitutes one of the core stages of the public policy cycle—alongside planning and programming—because it allocates resources according to policy priorities and directly influences service quality, macroeconomic outcomes, and overall welfare (Schick, 1998; Atiyas & Sayın, 1997; Çetinkaya et al., 2011). Although budgets are prepared annually in most countries, their economic effects extend beyond a single fiscal year, making the assessment of fiscal policy through one-year horizons inherently limited (Lazăr & Andrei, 2006; Filc & Scartascini, 2009). Since expenditure decisions often have multi-year implications, aligning spending with available resources over the medium term has become essential (PEFA, 2011). Consequently, many governments began transitioning from traditional annual budgeting to medium-term budgeting (MTB) frameworks beginning in the mid-20th century.

Providing multi-year revenue and expenditure projections alongside the annual budget constitutes a fundamental form of MTB (Harris et al., 2013). Although initially adopted mainly in developed economies, the conceptual roots of MTB date back to the development-planning efforts of the 1950s, when the limitations of annual budgets in responding to economic shocks became evident (Schiavo-Campo, 2009; Kesik, 2005). These constraints, combined with the economic crises of the period and the shifting views on government roles, motivated OECD countries to adopt MTB practices more widely from the early 1970s onward (Diamond, 2006; Allen & Tommasi, 2001).

Multi-year expenditure forecasts enable governments to estimate the cost of maintaining service levels and link spending decisions with available resources, thereby improving the prioritization of public projects and ensuring a feasible medium-term horizon (Schroeder, 2007; Jena, 2006). In this context, medium-term revenue forecasts are particularly critical. The literature consistently shows that forecast errors increase as the horizon lengthens (Penner, 2001), intensifying fiscal uncertainty and the risk of resource misallocation. Accurate medium-term forecasts support fiscal discipline, enhance the continuity of public services, and allow for more informed timing and compositions of expenditures.

Significant forecast deviations may disrupt the balance between expected and actual revenues, forcing governments to cut spending or borrow unexpectedly, with potential consequences for service quality and social welfare. As global economic uncertainty has intensified, both developed and developing countries have increasingly emphasized research on improving revenue forecasting methods and enhancing the reliability of fiscal projections. Such efforts provide valuable information for decision-makers by incorporating not only historical patterns but also real-time economic indicators.

The economic literature continues to debate the broader implications of MTB. Some authors argue that MTB better incorporates intergenerational concerns and offers a more political representation of the future (White, 2018), while others view it as an instrument for reconciling short- and long-term fiscal needs (Schack, 2009). Nonetheless, there is broad consensus that MTB offers several key advantages: It strengthens long-term planning and protects against inefficient spending (Muzychenko et al., 2017), links annual budgets to strategic priorities, and helps policymakers anticipate policy lags. It also improves managerial flexibility and the early identification of emerging fiscal issues (Harris et al., 2013). Furthermore, MTB enhances accountability through regular reviews of commitments and the integration of medium-term audit mechanisms (Boex et al., 2000).

International organizations—including the OECD, IMF, World Bank, and European Union—have extensively studied MTB, documenting its contribution to fiscal sustainability, economic stability, and improved public financial management. Growing empirical evidence suggests that the adoption of MTB can strengthen efficiency and support the broader realization of economic and social policy objectives.

1.2. Literature Review and Hypotheses

Research in the field has pointed out that the accuracy and objectivity of public budget forecasts depend on the interval of time. Examples of such studies include Gentry (1989), Larkey and Smith (1989), Mayper et al. (1991), Reddick (2008), Voorhees (2006), Calabrese and Williams (2019), Lago-Peñas and Lago-Peñas (2008), Benito et al. (2015), Ríos et al. (2018), Geys et al. (2008), Sedmihradská and Čabla (2013), Boukari and Veiga (2018), Lee and Kwak (2020), McShea and Cordes (2019), Bischoff and Gohout (2006), Buettner and Kauder (2015), Kauder et al. (2017), Jochimsen and Lehmann (2017), ElBerry and Goeminne (2020), and Picchio and Santolini (2020). The vast majority of the studies scrutinize yearly budget forecast errors, while the studies solely dedicated to the exploration of the causes of medium-term budget forecast errors continue to be few in number. Examples of such studies include Plesko (1988), Heinemann (2006), Frankel (2011), Breuer (2015), Harris et al. (2013), Allen et al. (2017), and Yılmaz (2019).

Most of these studies have found that revenue forecasts exhibit a negative bias, meaning that the actual revenues fall short of the forecasted figures. This study analyzes inaccuracy and bias by calculating the Forecasting Error (FE), Mean Percentage Error (MPE), Mean Absolute Percentage Error (MAPE), Number of Negative Error Periods (NoNEP), and Number of Positive Error Periods (NoPEP) and by conducting the Holden and Peel (1990) test.

Hypothesis 1.

The medium-term budgets are forecasted inaccurately.

Hypothesis 2.

The medium-term budget forecast errors are predominantly negative, indicating a systematic downward bias.

Gentry (1989), Mayper et al. (1991), Boukari and Veiga (2018), and Brogan (2012) found a correlation between unemployment and forecast errors. In contrast, Sedmihradská and Čabla (2013) and Afonso and Silva (2012) did not find a significant relationship. The theoretical expectation is that as unemployment rises, revenue forecast errors will trend negatively, as increased unemployment reduces household spending, leading to lower revenue collection.

Brogan (2012), Buettner and Kauder (2015), Sedmihradská and Čabla (2013), Merola and Pérez (2013), Boukari and Veiga (2018), and Kara (2024b) found a significant connection between GDP and forecast errors. Bağdigen (2002) has partly verified this correlation. On the contrary, Ríos et al. (2018) did not observe such a link. Drawing on these studies, we hypothesize actual GDP levels as a major factor causing the budget forecast errors of developed countries. After all, the GDP is a direct determinant of revenue collection and economic activity; thus, the variances in the actual GDP during the specified period have a huge impact on the sizes of the errors in budget forecasts.

Gentry (1989), Mayper et al. (1991), Lago-Peñas and Lago-Peñas (2008), Benito et al. (2015), and Deus and de Mendonça (2017) have identified a connection between population size and forecast errors. Therefore, we predict that the rates of population growth will have a major impact on budget forecasting errors. Since population growth is a factor that affects both economic demand and government spending, non-conformities in the actual population growth rate are going to be the reason behind the discrepancies in budget forecasts.

Hypothesis 3.

Macroeconomic conditions systematically influence medium-term revenue forecast errors.

We incorporate two structural fiscal indicators, the current account balance as a percentage of the GDP and the general government final consumption expenditure as a percentage of the GDP, into a unified hypothesis, as both variables capture fundamental aspects of a country’s fiscal and macroeconomic position. The current account balance reflects the scale of external imbalances arising from trade and financial flows, which can transmit volatility to fiscal aggregates and complicate revenue projections. Significant deviations in this indicator may therefore contribute to forecast inaccuracies by altering the external environment in which fiscal policy operates.

Similarly, the general government final consumption expenditure measures the level of spending on goods and services relative to the size of the economy. Shifts in this expenditure ratio can affect budget forecasts by influencing the trajectory of fiscal outcomes, particularly when changes in government consumption modify the demand structure of the economy or alter the composition of public spending. Taken together, these two indicators reflect core elements of fiscal structure and institutional capacity that are expected to influence the accuracy of medium-term revenue forecasts.

Hypothesis 4.

Fiscal structure and institutional capacity affect the accuracy of medium-term revenue forecasts.

Fiscal rules limit government borrowing, spending, and debt, potentially improving fiscal discipline, thereby reducing forecast errors. In particular, those budget balance rules aiming to contain deficits focus on ensuring fiscal discipline so that public finances remain stable. Fiscal rules may lower forecast errors by ensuring conservative fiscal management but could also lead to divergence when conditions evolve in an unexpected way, thus limiting the space for possible adjustments in budgets. Several studies are known to have established a significant relationship between fiscal rules and budget forecast errors, including Smith (2007), Beetsma et al. (2009, 2011, 2013), Pina and Venes (2011), Chatagny (2015), Afonso and Silva (2012), Hagen (2010), Frankel (2011), Holm-Hadulla et al. (2012), Baldi (2016), Heinemann (2006), Strauch et al. (2004), Chakraborty and Sinha (2008), and Picchio and Santolini (2020). Therefore, based on these findings, we hypothesize that the adoption of fiscal rules has a prudent influence on budget forecast errors.

Hypothesis 5.

Fiscal rules influence budget forecast errors.

In times of economic crisis, the level of uncertainty becomes very high, and this, in turn, leads to the volatility of the main economic indicators. The instability caused by this situation may be one of the reasons that forecast errors may increase, since governments will probably have a hard time revising their budgetary projections in a much quicker manner to adapt to rapidly changing situations. A number of different researchers have established a close link between crisis periods and government budget forecast errors, as seen in the works of Pew Center on the States and The Nelson A. Rockefeller Institute of Government (2011), Calitz et al. (2013), Boukari and Veiga (2018), and Kara (2024b), among others. In light of these conclusions, we put forward the hypothesis that crisis periods undermine the accuracy of budget forecasts.

Hypothesis 6.

Economic crisis periods influence budget forecast errors.

1.3. Dataset and Methodology

The research covers the medium-term budget projections of eight advanced countries (Australia, Canada, Germany, Hong Kong, Ireland, New Zealand, the UK, and the USA), with a special emphasis on two- to three-year budget projections. More precisely, when the budget for the year (t) is made in the year (t − 1), forecasts for the next year (t + 1) and the following year (t + 2) are also prepared. As far as this analysis is concerned, the forecasting for the year (t + 1) is called the “out-year,” and the forecasting for the year (t + 2) is called the “outer-year.” The reviewed period is from 2009 to 2023. However, if we take into account the period when the forecasts were actually prepared, the pertinent period is 2007 to 2021.

The decision to include the eight selected countries was grounded in their long-term transition to medium-term budgeting (MTB) practices, which enabled the performance of a thorough analysis. In addition, the forecasts for their medium-term budgets were calculated according to the same basic methods, which made the study comparison possible (Harris et al., 2013; Kara, 2024a; Moretti et al., 2023; Beetsma et al., 2013; Beetsma et al., 2011; Bischoff & Gohout, 2006; Boukari & Veiga, 2018; Breuer, 2015; Calitz et al., 2013).

It is very necessary to specify the extent and limits of this research. This study will not evaluate the particular methods, structural components, or institutional frameworks that are the basis for the budget forecasting in each country. While these factors are relevant, their analysis is not included in this study, which is rather ongoing research that is already complementary to this one. This study’s main objective is the identification and analysis of forecast errors in a group of comparable countries that were selected because of their similarity in MTB practices.

This intentional option makes it so that differences in the production of forecasts do not spoil the comparability of the results, since the countries that were chosen have very similar methodologies. This study thus escapes the introduction of biases or confounding variables due to the use of vastly different forecasting methods by completely isolating these factors in the analysis. Furthermore, the period that was chosen represents all countries’ common intersection period and thus allows the assessment of their budget forecasts and outcomes to be performed uniformly. The limits set here have been clearly stated to make it easier to conduct an investigation into forecast errors without having to undertake wider methodological or institutional analyses, which are beyond the intended scope of this study.

$$\mathrm{F}{\mathrm{E}}_{\mathrm{t}}={\displaystyle \frac{\left({\mathrm{A}}_{\mathrm{t}}-{\mathrm{F}}_{\mathrm{t}}\right)\ast 100}{{\mathrm{F}}_{\mathrm{t}}}}$$

(1)

$${\mathrm{M}\mathrm{P}\mathrm{E}}_{\mathrm{t}}={\displaystyle \frac{1}{\mathrm{T}}}{\sum }_{\mathrm{T}=1}^{\mathrm{T}}{\displaystyle \frac{\left({\mathrm{A}}_{\mathrm{t}}-{\mathrm{F}}_{\mathrm{t}}\right)\ast 100}{{\mathrm{F}}_{\mathrm{t}}}}$$

(2)

$${\mathrm{M}\mathrm{A}\mathrm{P}\mathrm{E}}_{\mathrm{t}}={\displaystyle \frac{1}{\mathrm{T}}}{\sum }_{\mathrm{T}=1}^{\mathrm{T}}{\displaystyle \frac{\left|\left({\mathrm{A}}_{\mathrm{t}}-{\mathrm{F}}_{\mathrm{t}}\right)\right|\ast 100}{\left|{\mathrm{F}}_{\mathrm{t}}\right|}}$$

(3)

The FE refers to the measurement of the deviation between the predicted and actual figures represented in terms of the percentage of the forecasted figure, which gives the user a direct measure of the magnitude of the error in relation to the actual values. Nonetheless, the FE may lead to incorrect conclusions when actual values are low, since the percentage error could become very high as a result of being out of proportion. The MPE finds the mean of all the percent errors of a particular forecast, thereby giving the user a picture of the overall bias of the predictions. Still, the MPE might be influenced by the sign of the errors, which might lead to the positive and negative errors being canceled out and to a view that is not very accurate given that the errors are not evenly distributed. The MAPE finds the average of the absolute values of percent errors, thereby giving a measure of the forecast accuracy that is not affected by the direction of the errors. The MAPE’s interpretability and simplicity are the main reasons behind its widespread adoption. However, it is still the case that its application in cases of zero or very low actual values suffers from the problem of generating disproportionately high errors. In the cases of advanced countries, forecast errors frequently result in values that are almost zero, whether they are positive or negative. Hence, the preference for using the MPE instead of the MAPE is justified for this study.

While the MPE provided insights into the possible bias in the forecasts, it was not able to tell whether the errors were constant and systematic over time or whether the bias was statistically significant. In order to solve this, the Holden and Peel (1990) test was used to detect the forecast bias. First, the MPE was linearly transformed to obtain the ME value. After that, a regression analysis according to Equation (4) below was conducted. A t-test (H0: λ = 0) was then used to test for the existence of bias in the forecasts at this point:

$$A_t-F_t=\lambda +U_t$$

(4)

The bias coefficient (λ) in the equation is similar to the magnitude of the ME, but it enables the investigation of whether the over- or underestimation of revenues is statistically significant. A positive and significant λ means that budget revenues are regularly underestimated, whereas a negative and significant λ posits the opposite.

Later, taking into account their existence and mutual correlations, the literature on forecast errors, and the theoretical framework, the economic variables were grouped. Below, the table presents comprehensive details about the data in terms of the variable acronyms and their sources (

Table 1. Dataset information.

Variable	Explanation	Acronym	Source
Out-Year	Out-year (t + 1) budget revenue forecasting error	Out-Year	National budgets
Outer-Year	Outer-year (t + 2) budget revenue forecasting error	Outer-Year	National budgets
Inflation	Consumer price inflation	INF	World Bank database
Gross Domestic Product	GDP as current USD	GDP	World Bank database
Unemployment	National unemployment rate	UNP	World Bank database
Population Growth	Population growth rate	PPG	World Bank database
Current Account Balance	Current account balance as % of GDP	CAB	World Bank database
Expenditure	General government expenditure as % of GDP	EXP	World Bank database
Budget Balance Rule	“1” if there is a budget balance rule in place; “0” otherwise	BBR	IMF Fiscal Rule database
Number of Numerical Rules	Total number of numerical rules in place	NNR	IMF Fiscal Rule database
Economic Crisis	“1” if there is an economic cris; “0” otherwise	CRS	The literature

).

The variables shown in the Table 1 were picked taking into account their importance in revealing the drivers of medium-term budget revenue forecast errors in advanced economies. We used normal values to capture the directionality of these errors, thereby permitting the examination of negative errors, which is the opposite of what studies usually perform when they concentrate on absolute forecast errors. This process makes the picture clearer as to whether revenue forecasts are always over- or under-forecasted. For the forecasts, the source is the medium-term budgets that are part and parcel of national budget laws instead of supplementary documents like medium-term programs or fiscal plans. This means that the forecasts that directly affect the fiscal policy are the ones being used.

As for all the other variables, the used values are for the years to be forecasted and not for the years when the forecasts were made. For example, in the analysis of the out-year forecast error, which is year (t + 1), the economic variables in the model also refer to (t + 1). This choice of methodology guarantees consistency because it links the forecast errors directly to the economic conditions present during the forecast year. Thus, the analysis captures the effect of the macroeconomic conditions that were realized during the target year on the precision of the forecasts.

The main idea of this method is to find out the answer to the following question: how much do forecasts reflect the economic situation of the target year accurately or, alternatively, are they only based on the data that was available at the time? This study assumes that the differences between the forecasted and actual macroeconomic variables (e.g., GDP) are one of the main factors that influence the results, but it still captures the economic situation in the target year itself (e.g., t + 1 or t + 2) as the main source of influence.

This study’s design supports its goal of specifying the interplay between the actual circumstances and the correctness of the forecast during the target time, rather than mixing this analysis with the different but related issue of the conditions of forecast preparation. The inclusion of realized economic variables in this study grants a deeper examination of the forecast performance in the light of reality, thereby providing an idea of their responsiveness and robustness.

To make it clear, although the conditions of the time of preparation are taken into account and acknowledged as a significant area of inquiry, they are not a concern of the current study. Future research might be based on this study by bringing these factors into a larger framework; but, for now, the current study is intentionally limited to the specific question of how the macroeconomic realities in the forecasted year affect the errors of the forecasts.

To sum up, the addition of these variables is based on both theoretical and empirical reasoning. Inflation and GDP growth will be the two main factors contributing positively to forecast errors because revenue estimation is likely to become more difficult with higher nominal activity and price volatility, a result corroborated by Aizenman and Hausmann (2000) and Frankel (2011). Unemployment and the quantity of fiscal rules are, however, expected to have a different impact, namely, a negative effect, since economic downturns usually go along with conservative revenue forecasts, and stronger fiscal frameworks promote discipline and prudence (Afonso & Silva, 2012; Beetsma et al., 2013). The population growth rate is likely to have a weakly negative or neutral impact, in that demographic changes may enhance the predictability of the tax base over a short period but lose their influence of clarification over a longer time span. Budget balance rules, although aimed at promoting fiscal discipline, might inadvertently lead to increased forecast errors when governments pursue excessive optimistic targets to comply with balance constraints (Chatagny, 2015; Baldi, 2016). The current account balance and expenditure ratio, together with crisis dummies, were also added as control variables, where their expected effects are not clear—external imbalances or crises may result in either under- or overestimation, depending on policy responses and institutional strength. The assumptions on directionality were formed on the basis of fiscal forecasting theory and previous empirical findings and thus provide a consistent rationale for the model specification.

The reasons for including each economic variable are explained below:

Inflation: Inflation has been included in the analysis because it has a direct effect on nominal revenue. If not correctly considered, higher inflation may result in the overestimation of revenue, since it enlarges the base on which forecasts are made. However, low-inflation periods can cause revenue to be underestimated. This economic factor is usually discussed in the literature, where it is exclusively related to forecast volatility and uncertainty.

Gross Domestic Product: The GDP is the best measure of a country’s economic activities and its main indicator of fiscal health. A poor GDP performance might result in the overly optimistic forecasting of revenues, while the opposite situation might lead to underestimation in times of economic growth. We selected the current GDP rather than GDP expectations because it more accurately depicts the actual impact of economic conditions on budget forecast errors.

Unemployment: Unemployment figures provide an indication of the labor market’s condition. A rise in unemployment usually means a fall in revenue from personal income tax and consumption-based taxes, increasing the possibility of an overly optimistic situation in revenue forecasting. One reason for including the unemployment variable is to gauge how much labor market conditions contribute to forecasting errors.

Population Growth: This variable was incorporated to account for demographic characteristics that influence the revenue generation and spending needs of the population. Countries with rapidly increasing populations may face considerable difficulties in forecasting revenues accurately due to shifting patterns of consumption and tax bases. There have also been contradictory findings from previous studies on the role of changes in population in forecasting errors, which makes it pertinent to study this variable further.

Current Account Balance: This is a new variable in the investigation of budget forecast errors, and we incorporated it to see whether any external imbalances, trade deficits for instance, might influence the prediction. Countries with persistent current account deficits may wish to see more volatility in revenues, leading to oversized errors in reporting economic or historical estimates.

Expenditure: The general government final consumption expenditure as a ratio of the GDP was chosen to help understand how priorities in government spending align with revenue forecasts. Changes in patterns of government expenditure, particularly in response to unforeseen, large, external economic shocks or changes in fiscal policy, can adversely affect the accuracy of forecasted revenues.

Budget Balance Rule: This dummy variable, reflecting whether or not there is a binding budget balance rule, was included because fiscal rules are intended to provide discipline to budget forecasts. Countries with strong fiscal rules are expected to behave with more caution in their revenue forecasts, given their concern with adhering to legal constraints.

Number of Numerical Rules: Similar to the budget balance rule, this variable captures the extent of fiscal constraints in place. The numerical rules that are higher in number might be indicative of greater fiscal discipline, which, in turn, could lead to less errors in forecasting.

Economic Crisis: Economic crises are taken into account because they usually result in huge forecast errors due to their unpredictable nature. Revenue falling short of expectation usually brings about significant discrepancies between original forecasts and actual revenue amounts. Thus, this variable is very important in recognizing extreme situations when forecasts may fail.

This study primarily aims to reveal the main economic factors in advanced economies that lead to budget forecast errors by concentrating on medium-term budget forecasts and introducing both already-known and new variables using the workflow shown in Figure 1. The application of real-time variables guarantees that the analysis corresponds to the actual circumstances under which the forecasts are made, thereby giving insights into the difficulties of keeping fiscal projections accurate.

The potential endogeneity issues, like the interrelation of unobserved country-specific effects with the independent variables, were meticulously investigated. The Hausman (1978) test was carried out on both models to determine whether fixed or random effects were applicable. The outcomes (p = 0.1402 for out-year and p = 0.2330 for outer-year) show that the hypothesis of no systematic correlation cannot be rejected, proving the random-effect specification to be correct. This implies that the endogeneity due to the non-inclusion of country-specific factors does not influence the estimators. In addition, to further cement the robustness, Driscoll and Kraay (1998) standard errors were used to lessen the impact of the biases caused by cross-sectional dependence and heteroskedasticity. These methods, while making our estimates more reliable, still allow for some forms of endogeneity, like reverse causation, to exist. Future studies could rely on the use of instrumental variables or the exploitation of natural experiments for more solid causal inferences.

While we were performing the estimation, we spotted the issue of heteroskedasticity in the data (Prob > chi² = 0.0000) (refer to

Table A6. Modified Wald test for heteroskedasticity.

Out-Year		Outer-Year
chi2(15)	98.74	chi2(15)	91.77
Prob > chi2	0.0000	Prob > chi2	0.0000

). These econometric problems can have an impact on the outcome, as heteroskedasticity makes the estimates inefficient, while cross-sectional dependence can result in the standard errors being underestimated, thereby increasing the risk of obtaining false positives.

We resolved these problems through the application of the standard error method of Driscoll and Kraay (1998). The method involved the adjusting of the standard errors with respect to both heteroskedasticity and cross-sectional dependence; this is what made it suitable for the panel data that covered several countries and periods of time. Traditional estimators, which assume that there is no cross-sectional correlation, are not similar to the Driscoll and Kraay (1998) method, which is still applicable to any form of cross-sectional dependence. It employs non-parametric techniques to cope with time and unit correlations, ensuring that the estimates are not only consistent but also reliable. The method also lessens the biases caused by heteroskedasticity, thereby allowing one to make stronger statistical inferences when comparing the budget forecast errors of different countries and periods repeatedly.

Here are the equations that were applied in the panel models for the out-year and outer-year:

$$\begin{array}{cc}\hfill Out-Yea{r}_{\left\{i(t+1)\right\}}& =\alpha +{\beta }_{1}IN{F}_{\left\{it+1\right\}}+{\beta }_{2}GD{P}_{\left\{it+1\right\}}+{\beta }_{3}UN{P}_{\left\{it+1\right\}}+{\beta }_{4}PP{G}_{\left\{it+1\right\}}+{\beta }_{5}CA{B}_{\left\{it+1\right\}}+\hfill \\ & {\beta }_{6}EX{P}_{\left\{it+1\right\}}+{\beta }_{7}BB{R}_{\left\{it+1\right\}}+{\beta }_{8}NN{R}_{\left\{it+1\right\}}+{\beta }_{9}CR{S}_{\left\{it+1\right\}}+{\epsilon }_{\left\{it\right\}}\hfill \end{array}$$

(5)

$$\begin{array}{cc}\hfill Outer-Yea{r}_{\left\{i(t+2)\right\}}& =\alpha +{\beta }_{1}IN{F}_{\left\{it+2\right\}}+{\beta }_{2}GD{P}_{\left\{it+2\right\}}+{\beta }_{3}UN{P}_{\left\{it+2\right\}}+{\beta }_{4}PP{G}_{\left\{it+2\right\}}+{\beta }_{5}CA{B}_{\left\{it+2\right\}}+\hfill \\ & {\beta }_{6}EX{P}_{\left\{it+2\right\}}+{\beta }_{7}BB{R}_{\left\{it+2\right\}}+{\beta }_{8}NN{R}_{\left\{it+2\right\}}+{\beta }_{9}CR{S}_{\left\{it+2\right\}}+{\epsilon }_{\left\{it\right\}}\hfill \end{array}$$

(6)

In both models, the constant term $\alpha$ is the one that has the universal effect when the independent variables are zero. The coefficients β₀, β₁, β₂, …, β₉ show how big the effect of an independent variable is on the dependent variable, and in which direction it goes. The error term (${\epsilon }_{\left\{it\right\}}$) represents the unobserved factors and random deviations from the model and was adjusted using Driscoll and Kraay standard errors (Hoechle, 2007).

The main reason for employing panel analysis in this research is to point out the structural similarities that cause medium-term revenue forecast errors in advanced economies, thereby providing hints that are relevant for developing countries that are in the process of adopting similar budgeting practices. It is true that country-specific studies might produce results that are only applicable to the corresponding institutional or fiscal context of that specific nation. Nevertheless, the panel approach helps to identify the shared dynamics and prove statistically significant relationships among countries with similar fiscal frameworks. Consequently, the countries chosen were Australia, Canada, Germany, Hong Kong, Ireland, New Zealand, the United Kingdom, and the United States due to their long histories of using medium-term budgeting systems, the presence of consistent and trustworthy data, and the application of forecasting methodologies that are at least somewhat similar. The joint analysis of these countries not only provides a clearer comprehension of the factors causing forecast errors in mature fiscal systems but also reveals structural patterns and challenges that developing economies might face as they move up to similar fiscal sophistication levels. Thus, the panel methodology enhances the comparative aspect of the research and permits a systematic study of the universal factors affecting revenue forecast accuracy in distinct institutional settings.

2. Results

First, the table below (

Table 2. Bias test results.

Country	Forecast	MPE	MAPE	ME/λ	p-Value	NEP	PEP
Australia	Out-year	−51.524	103.696	−212.933	0.1024	12	3
	Outer-year	−51.411	106.618	−237.466	0.1064	11	4
Canada	Out-year	−0.3350	48.084	0.9866	0.8170	9	6
	Outer-year	−0.5568	75.219	16.600	0.8415	9	6
Germany	Out-year	0.6293	45.443	19.800	0.6834	3	12
	Outer-year	0.7285	53.999	19.400	0.7254	6	9
Hong Kong	Out-year	0.4010	81.225	−76.466	0.4927	7	8
	Outer-year	14.072	73.830	−13.066	0.8906	7	8
Ireland	Out-year	−42.860	82.866	−19.510.670	0.2921	8	7
	Outer-year	−53.472	125.277	−24.746.000	0.3899	8	7
New Zealand	Out-year	31.897	49.460	28.465.330	0.1300	8	7
	Outer-year	25.839	53.385	28.472.000	0.1532	6	9
The United Kingdom	Out-year	12.145	37.573	98.800	0.3569	3	12
	Outer-year	0.3128	51.602	57.866	0.6854	5	10
The United States	Out-year	−84.453	116.800	−2.591.267	** 0.0194	13	2
	Outer-year	−108.770	143.361	−3.505.867	*** 0.0064	12	3

MPE: Mean Percentage Error; MAPE: Mean Absolute Percentage Error; ME: Mean Error; Obs: observations; PEP: Number of Positive Error Periods; NEP: Number of Negative Error Periods; (*) indicates MacKinnon (1996) one-sided p-values: (***) significance at the 1% level; (**) significance at the 5% level.

) presents the MPE, MAPE, NoNEP, NoPEP, and the results of the Holden and Peel (1990) test for each country individually.

Looking at the MAPE numbers in the Table 2, it is clear that the United Kingdom has the best forecasting accuracy among the countries being compared, with MAPE values of 3.7573 for the out-year and 5.1602 for the outer-year. The United States, however, appears to have the worst predictions, with MAPE figures of 11.6800 for the out-year and 14.3361 for the outer-year.

Taking the MPE numbers into account, once again, Germany and the UK stand out as the most accurate forecasting countries, while the US comes last in this regard, with Australia closely behind. Additionally, the study of the NEP and PEP values shows that Australia, Canada, and the US all exhibit the same trend by being overestimated, and this is indicated by the negative bias. In contrast, Germany and the UK seem to be experiencing the opposite situation of being underestimated and thus show a positive bias. The other countries, in turn, seem to have the occurrence of both negative and positive signs in their forecasting errors that balance each other out.

Yet, in order to reinforce the econometric importance of the findings, the Holden and Peel (1990) test was employed. The findings indicate that only the US showed a negative bias in its predictions, with the p-values for the out-year (p = 0.0194) and outer-year (p = 0.0064) both pointing to a significant overestimation. This suggests that the prevalence of both negative and positive signs in forecasts from other countries is not indicative of bias but rather results from a variety of economic, political, structural, institutional, and technical factors.

The reasons for the errors of the out-year budget revenue forecasts are presented in the

Table 3. Out-year model results.

	Coef.	Drisc/Kraay Std. Err.	t	P > t	[95% Conf. Interval]
INF	0.0228	0.0026	8.68	*** 0.000	0.0166	0.0291
UNP	−0.0100	0.0037	−2.70	** 0.031	−0.0188	−0.0012
PPG	−0.0177	0.0090	−1.96	* 0.091	−0.0390	0.0036
BBR	0.0332	0.0111	2.98	** 0.021	0.0068	0.0597
NNR	−0.0249	0.0042	−5.82	*** 0.001	−0.0351	−0.0148
CRS	−0.0315	0.0193	−1.63	0.147	−0.0772	0.0142
GDP	0.0001	0.0000	−4.43	*** 0.003	0.0000	0.0000
CAB	−0.0006	0.0015	−0.45	0.665	−0.0042	0.0028
EXP	0.0008	0.0009	0.91	0.395	−0.0014	0.0032
cons	0.0355	0.0383	0.93	0.386	−0.0552	0.1262
Number of obs.			120	Prob > F		0.0000
Number of groups			15	R-squared		0.4477
F (9. 7)			330.53	Root MSE		0.0751

(*) MacKinnon (1996) one-sided p-values: (***) significance at the 1%; (**) at the 5% and (*) at the %10 level.

.

Analysis results in the Table 3 show that the INF, NNR, and GDP are significant at the 1% level, the UNP and BBR are significant at the 5% level, and the PPG is significant at the 10% level. The results show that higher inflation rates increase out-year forecast errors. This finding is consistent with the existing literature on the negative effects of inflation on budget forecasting, particularly its consequences on prices, revenue collection, and economic expectations. Similarly, the positive effect of the GDP suggests that forecast errors increase during periods of higher economic growth, as governments overestimate future revenues.

The NNR has a negative and statistically significant effect on forecast errors. This suggests that countries with more stringent fiscal rules tend to make more accurate revenue forecasts. Such rules may reinforce fiscal discipline and, therefore, reduce upward bias in revenue forecasts. Unemployment also has a negative effect on forecast errors, suggesting that when unemployment is high, governments make more conservative revenue forecasts, possibly to hedge against any decline in revenues.

The BBR increases forecast errors, suggesting that such rules lead to overly optimistic revenue forecasts to maintain fiscal discipline, either because of political pressures or over-reliance on balanced budget targets. Finally, the PPG decreases forecast errors, suggesting that countries with growing populations can forecast future revenues more precisely, either because they can use better tax collection mechanisms or because of more sophisticated forecast models considering demographic change.

Following the results of out-year model, the results of the outer-year model are shown in the table below (

Table 4. Outer-year model results.

	Coef.	Drisc/Kraay Std. Err.	t	P > t	[95% Conf. Interval]
INF	0.0242	0.0028	8.41	*** 0.000	0.0174	0.0311
UNP	−0.0187	0.0034	−5.39	*** 0.001	−0.0270	−0.0105
PPG	0.0031	0.0125	0.25	0.810	−0.0266	0.0329
BBR	0.0608	0.0089	6.76	*** 0.000	0.0395	0.0820
NNR	−0.0360	0.0039	−9.18	*** 0.000	−0.0452	−0.0267
CRS	−0.0342	0.0194	−1.76	0.122	−0.0803	0.0118
GDP	0.0001	0.0000	−6.69	*** 0.000	0.0000	0.0000
CAB	−0.0004	0.0014	−0.31	0.765	−0.0039	0.0030
EXP	−0.0007	0.0010	−0.72	0.496	−0.0032	0.0017
cons	0.0943	0.0364	2.59	0.036	0.0081	0.1804
Number of obs.			120	Prob > F		0.0000
Number of groups			15	R-squared		0.5445
F (9. 7)			788.67	Root MSE		0.0818

(*) MacKinnon (1996) one-sided p-values: (***) significance at the 1% level.

).

Analysis results in the Table 4 show that the INF, UNP, BBR, NNR, and GDP are significant at the 1% level. The fact that all these variables are significant at the 1% level indicates that the importance of these factors has increased in outer-year revenue forecasts. As in the out-year model, the INF, BBR, and GDP have a positive effect on forecast errors, indicating that higher inflation rates and GDP levels and strict budget balance rules lead to larger errors in outer-year forecasts. The UNP and NNR, in contrast, decrease forecast errors, which indicates that raising the unemployment rate, along with the application of more numeric rules, leads to the higher accuracy of the forecast and, thus, smaller errors.

Though the PPG is an important factor in the out-year model, its statistical significance fades away in the outer-year analysis. This indicates that population growth is a factor in one-year budget forecasts, but its effect lessens in longer-term forecasts, perhaps because of the lagged impacts of demographic changes on fiscal conditions.

The next table shows up when we take into consideration the results of the analysis of both models (

Table 5. Significance results for both models.

Variable	Out-Year Model	Outer-Year Model
INF	Affecting (+)	Affecting (+)
UNP	Affecting (−)	Affecting (−)
PPG	Affecting (−)	Ineffective
BBR	Affecting (+)	Affecting (+)
NNR	Affecting (−)	Affecting (−)
CRS	Ineffective	Ineffective
GDP	Affecting (+)	Affecting (+)
CAB	Ineffective	Ineffective
EXP	Ineffective	Ineffective

).

The general evaluation of the variables that influence the budget revenue forecast error in both models can be seen in the Table 5. It is clear from the table that the significant findings of variables such as the INF, UNP, BBR, NNR, and GDP in both models suggest that these factors constitute the main causes of medium-term budget revenue forecast errors. This finding points out that both two-year and three-year budget forecasts rely importantly on these economic and fiscal variables. Meanwhile, variables such as the CRS, CAB, and EXP failed to affect either of the two models significantly. This shows that such factors are not determinant of budget forecast errors for the analyzed countries and periods and do not affect the forecast accuracy. The PPG showed its significance only in the out-year model and did not reveal any effect on the outer-year model, indicating that demographic changes are more powerful in two-year periods but do not play a decisive role in forecast errors in three-year periods. Generally, this assessment may point out that the main causes of budget revenue forecast errors are limited to some variables while other factors are less important in MTB.

3. Discussion and Policy Recommendations

The examination of errors in budget revenue forecasts in both the out-year and outer-year models provides essential information about the economic and fiscal factors that influence the correctness of budget forecasts. The findings show that a number of main variables (INF, UNP, BBR, NNR, GDP) have a significant effect on the errors in forecasting budget revenues in both models (see

Table 6. Hypothesis results.

No	Hypothesis	Status
		Out-Year	Outer-Year
1	The medium-term budgets are forecasted inaccurately.	Mostly rejected	Mostly accepted
2	The medium-term budget forecast errors have a systematic downward bias.	Rejected, except for the USA	Rejected, Except for the USA
3	Macroeconomic conditions systematically influence medium-term revenue forecast errors.	Accepted, except for the population growth rate	Accepted, except for the population growth rate
4	Fiscal structure and institutional capacity affect the accuracy of medium-term revenue forecasts.	Rejected	Rejected
5	Fiscal rules influence budget forecast errors.	Accepted	Accepted
6	Economic crisis periods influence budget forecast errors.	Rejected	Rejected

). The constant significance of these variables emphasizes their basic role in the fiscal decision-making processes of the developed countries. In particular, the positive effect of inflation and the GDP on forecasting errors implies that rising prices and economic growth may contribute to the uncertainty of revenue forecasts, forcing authorities to be more careful in their fiscal plans. This result is in line with the wider literature on fiscal forecasting, which points out the requirement of flexibility in budgetary practices in reaction to economic changes.

The inverse correlation between unemployment and forecasting errors underscores the inclination of higher-unemployment-rate governments to use more conservative revenue forecasts. This conduct is most probably a result of the realization that economic declines lead to lower tax revenues, which, in turn, cause governments to be less optimistic about revenue to avoid the risk of budgetary deficits. Thus, policymakers should be mindful of this at the time of the making of budgetary frameworks and should take care to provide sufficient support for the local authorities during economic downturns in order to maintain fiscal stability.

The PPG, as it were, was not obtained in the outer-year model, while it was evidently of great importance in the out-year model. This contrast points to the likelihood that demographic changes might have an even stronger impact on short-term forecasting, perhaps because of the need to cope with the fact that growing populations are a more pressing issue. The decision-makers are to carry out extensive data collection and, along with it, to take very comprehensive analyses on the topic of population movements, as knowledge of these trends might be the key to the accuracy of revenue forecasts.

Notwithstanding the literature that points out a strong link between crisis periods and budget forecast errors, this investigation reported no such relationship for crisis periods in the t + 1 and t + 2 models. One of the reasons for the difference in findings could be the method of analysis which, in this case, was less dependent on the assumptions made during the forecast formulation and more dependent on the actual macroeconomic conditions. The countries included in this study might also possess institutional measures (like automatic stabilizers or conservative forecasting practices) that reduce the crisis’s direct effect on the forecast accuracy. This paper could be the starting point of a larger narrative in which the topic of potential differences in institutional or methodological factors that separate this sample from the major patterns in the literature could be explored.

The other variables, like the CAB and EXP, were found to be very insignificant, which implies that they are not very involved in the budget forecasting accuracy in the countries where the analysis was conducted. Hence, decision-makers should rethink the importance attached to these variables during budgetary processes, as it may be possible that some of the indicators do not directly influence forecast outcomes. The scenario might then turn out to be better organized and more efficient in terms of budget allocation, as the process would be centered on the variables with the most significant impact.

The conclusions of this study have important implications for suggested policy measures. For the improved accuracy of budget forecasts, especially with a medium-term focus, given their importance to effective medium-term planning, the following steps need to be considered by policymakers:

Enhanced Training and Capacity Building: Investments in the development of budget officials and analysts must be made through training on the complexities of economic variables and how they affect forecasting. A more complete level of understanding of the relationship between inflation, GDP, and unemployment with budget projections will, to a larger degree, enable accepted decision making.

Adoption of Dynamic Forecasting Models: Usually, the old-fashioned methods of forecasting are unable to deal with modern economies’ complexities thoroughly. Instead, the new era in forecasting presents dynamic modeling that can either push towards machine learning or artificial intelligence. By this method, real-time data can be included so that the making of decisions is based on the information that actually reflects the macroeconomic conditions as they change. The skill of detecting intricate patterns in the data, together with the continuous routine of updating the forecasts, takes medium-term planning as the most suitable application. If methods of this caliber are used, forecasting could turn out to be very responsive and accurate in the direction of managing macroeconomic uncertainties.

Regular Review of Budgeting Processes: Economists may also have to set up mechanisms for the continuous assessment of budgetary procedures. This would imply recurrent reviewing to be able to adapt the assessment of how well the current practices actually work and to outline any changes that correspond to the economic indicators emerging. Thus, the budget process is continuously and automatically adjusted according to economic climate fluctuations, which is, in fact, the intention behind this setup.

Collaboration Between Economists and Policymakers: The strengthening of partnerships between economists and budget officials will have them practicing not only forecasting but also the integration of economic insights into the budgeting process, which will guarantee that forecasts are based on sound economic theory and empirical evidence. This partnership is necessary for proper medium-term planning, as it provides decision-makers with the power of informed choices considering a large variety of economic conditions.

Greater Transparency and Communication: The main way to improve the weak credibility of forecasts could be through stakeholders’ increased budget raising based on transparent and sound assumptions and methodologies. Public participation and engagement with the stakeholders in the budget process could also provide more informed feedback and expectations.

Scenario Planning and Sensitivity Analysis: The introduction of scenario planning and sensitivity analysis would shed light on the different possible outcomes under altered economic conditions for policymakers. This will enable them to apply a wider system of long-term planning, thereby making it easier for the government to be ready for various economic fluctuations and synchronize its budget strategy accordingly.

Strengthening Legal and Institutional Frameworks: The establishment or a considerable enhancement of the legal and institutional frameworks that require adherence to MTB principles will gradually build trust. Such a framework should include periodic reporting, measures of accountability, and comprehensive guidelines for the adjustments of budgets along with revenues received.

Technology and Data Analytics Utilization: Technology and superior data analytics, if fully capitalized, could put revenue forecasts in a much better position. Decision-makers need to pour money into systems that will enable the integration of big data and analytics into the budgeting process for the purpose of enhancing data-driven decision making.

Encouragement of Stakeholder Engagement in Budget Processes: Participation of different stakeholders, such as civil society organizations and the private sector, will not only increase the area’s diversity but will also give the discussion of budgets more general legitimation. Participation means that different points of view are involved in the process, and this is particularly true for developing more accurate and relevant forecasts.

The results obtained from the research, besides improving the methods, have implications for fiscal institutions in practice. The continual effect of inflation, economic growth measured by GDP, and fiscal rules on the errors made in forecasting indicate the necessity of institutional arrangements that promote regular macroeconomic monitoring as part of the revenue estimation process. The creation of independent fiscal councils or bolstering the capability of existing forecasting units in finance ministries could both increase the quality of the analysis and limit the extent to which optimism influenced by political considerations could affect the outcome.

In addition, governments in different countries should think about adopting a post-forecast evaluation system that periodically measures the difference between the predicted and actual results. Such feedback channels would provide organizations with the opportunity to acquire knowledge from the errors they make systematically, change the parameters of the models, and improve the assumptions they hold over time. The learning-by-doing process of institutions may have a very positive effect on the accuracy and trustworthiness of revenue forecasts, particularly in the context of medium-term budgeting.

In conclusion, improving communication between statistical offices, tax authorities, and budget departments would facilitate the incorporation of up-to-the-minute data into fiscal projections in a very effective way. This move would lead to a dual benefit of more accurate forecasts and enhanced institutional capability for the very fiscal planning that is considered credible and thus sustainable.

Whether it is the out-year or outer-year model, several macroeconomic and fiscal aspects have been spotlighted, indicating their pivotal role in the accuracy of revenue predictions. It can be reasonably presumed that if governments adopt these measures, they would literally walk into the territory of higher accuracy in the forecasts of revenues and therefore more stable financial situations. This study accentuates the intricate dependency among the economic situation, budget forecasts, and policy, thereby signalizing the need for responsive and informed fiscal policy in the ever-changing economic environment. Focusing on medium-term planning helps not only to maintain fiscal discipline but also to prepare governments for future challenges that can be turned into opportunities.

4. Conclusions

Revenue forecasting for the medium term is the main pillar of good fiscal management and sound, strategic decision making in developed countries, where reliable projections serve as the foundation for long-term plans and fiscal stability. Unreliable forecasts could also result in budget deficits, the incorrect distribution of funds, or the interruption of public services. The present research aims at adding to the body of knowledge concerning the factors causing revenue forecast inaccuracies in different developed countries and thus provides understanding for the betterment of such forecasts’ accuracy.

The findings reveal a number of quite moderate aspects that cause errors in medium-term revenue forecasts. These include inflation and GDP because they simply increase forecast errors, whereas there are nearly correct projections of revenues in a very dynamic or rapidly growing economy. Certainly, this means that there is a lot of uncertainty in the economy, combined with various shocks that impact fiscal projections.

On the contrary, unemployment, the population growth rate, and the number of fiscal rules reduce forecast errors, indicating that conservative approaches might be adopted during uncertain phases for the economy, or stricter fiscal governance, that lead to a more prudent and thus accurate forecast. Yet, in particular, if there are budget balance rules, this may lead to larger forecast errors, since very stiff fiscal targets are set without an allowance for fluctuation in economic terms.

Another interesting finding of this study is that the significances of the variables are rather different in the out-year and outer-year models. In the outer-year model, the variable of population growth loses its significance, while all the other variables tend to gain stronger statistical significance. It can be concluded that forecast errors practically grow in the long run over the line distance, which, in turn, stresses the need for better tools and methodologies in the preparation of medium-term revenue forecasts, especially in the case of multi-year forecasting setups.

The results reveal the challenges involved in forecasting budgets for the medium-term period and stress the constant need to revisit the assumptions behind and the policies on the use of public funds. If these factors are properly managed, the revenue estimates of the government can be made more accurate, the likelihood of fiscal risks can be lessened, and the public financial management system would prove to be more effective in the medium term.