Analyzing the dynamic effects of Reserve Bank of Australia’s cash rate and oil price fluctuations on equity returns in major Australian companies Open Access

This study aims to empirically investigate the dynamic effects of changes in the cash rate and oil prices on the stock returns of Australia’s fifteen largest companies, which constitute over 50% of the ASX200 market capitalization, using a general-to-specific modeling approach. The significance of this study lies in its ability to provide a robust modeling framework that captures the intricate relationships between macroeconomic variables and stock returns, offering valuable insights for investors and policymakers. The primary research questions addressed in this study are:

To answer these research questions, the study uses a general-to-specific modeling strategy. This approach begins with a comprehensive general model that includes all relevant variables and iteratively simplifies it by removing statistically insignificant components. The final specific models are tested for robustness and validity using various diagnostic tests.

Given the fact that the Reserve Bank of Australia (RBA) will soon start cutting the cash rate, it is important to have an in-depth understanding of the effects of changes in interest rates on the largest Australian companies. This knowledge is crucial for investors and policymakers as interest rate adjustments can significantly influence economic activity, corporate profitability and stock market performance. Lower interest rates typically reduce borrowing costs, stimulate consumer spending and encourage business investments, but such dynamic impacts can vary widely across different stocks. Understanding these differential effects can help superannuation funds, institutional investors and retail investors with portfolio rebalancing strategies during periods of changing interest rates, oil prices and specific calendar months.

Previous studies have established that nominal interest rates can predict stock returns (see, among others, Campbell, 1987; Fama and French, 1989; Keim and Stambaugh, 1986; Ang and Bekaert, 2007; Reilly, Wright, and Johnson, 2007; Časta, 2023; Caporale, Gil-Alana and Melnicenco, 2024). According to Mishra et al. (2023), the interest rate channel represents the primary conduit through which monetary policy impacts the real economy. Mean shifts in US real interest rates, predominantly driven by factors such as budget deficits and oil shocks, exert significant influence on this channel. One could argue that this relationship extends to its impact on the equity market through its connection with the implementation of inflation targeting. Ratti and Vespignani (2019) show that global official interest rates are mainly driven by changes in global monetary aggregates, oil prices, global output and global prices. Conversely, the US adjusts its official interest rates based on both global output and inflation, using a framework akin to a Taylor rule.

Using a multifactor model, other studies (Faff and Chan, 1998; Faff and Brailsford, 1999; Sadorsky, 2001; Kilian and Park, 2009; Cong et al., 2008; Caporale et al., 2022; Gogineni, 2010) found that crude oil prices can also influence stock returns. A third group of studies emphasized the importance of capturing calendar anomalies, cyclical and seasonal movements in stock returns (e.g. Wachtel, 1942; Keim, 1983; Agrrawal and Skaves, 2015; Levy and Yagil, 2012; Chatzitzisi, Fountas, and Panagiotidis, 2021; Steinborn, 2024).

This study makes timely contributions to the existing literature by providing empirically validated insights into the effects of changes in the cash rate and oil prices on the stock returns of Australia’s fifteen largest companies, which collectively represent half of the ASX200 market capitalization. By using a theoretically-relevant set of macroeconomic factors within an unrestricted dynamic general model, the study captures the complex dynamic relationships between these variables and stock returns. The estimated models pass rigorous diagnostic tests, demonstrating superior performance relative to other general models, as indicated by model selection criteria such as the Akaike Information Criterion (AIC), Schwarz Information Criterion (SIC), and Hannan-Quinn Criterion (HQC). These findings offer a deeper understanding that is crucial for both investors and policymakers, enabling more informed decision-making in alignment with broader economic trends and stock-specific dynamics. For investors, the practical implications are substantial: superannuation funds, institutional investors and retail investors can use these insights to optimize portfolio allocations in response to shifting economic conditions. Specifically, in periods of rising interest rates, reallocating investments away from sectors most sensitive to borrowing costs and consumer spending could help mitigate potential losses. Similarly, in times of increasing oil prices, adjusting portfolios to reduce exposure to sectors with higher operational costs may enhance resilience.

In this paper, we adopt the General-to-Specific Modeling (GETS) methodology (Hendry, 1995, 2003), a modeling strategy developed as part of the London School of Economics (LSE) approach. The GETS method offers several key benefits, including a systematic approach to model selection, which begins with a general model and iteratively simplifies it by removing statistically insignificant or theoretically irrelevant variables (Hendry, 2003). This process helps prevent overfitting and ensures the final model is both parsimonious and robust, aligning with the parsimony principle. A successful modeling strategy, as outlined by Hendry (1995), should identify all key determinants and their lag dynamics, ensure valid conditioning with stable parameters and handle distributional shifts, outliers and fat-tailed distributions (Castle, Doornik, and Hendry, 2021).

The LSE approach views econometric models as approximations of an unknown data generation process (DGP). It emphasizes parsimony and validity, aiming to develop simple models that capture essential features without unnecessary complexity while reflecting economic theory and empirical evidence. The GETS method also captures dynamic responses effectively by incorporating rigorous diagnostic testing at each step, improving model reliability and validity. Furthermore, by focusing on the most relevant variables, the GETS approach enhances model interpretability and practical relevance in empirical research. This method is particularly effective in producing well-specified, robust econometric models that accurately capture key dynamic responses and is now available in the EViews software program. Hendry (1995) defines econometrics as testing, testing and testing where theory guides the identification of relevant factors, and then empirical data support or refute those theoretical relationships. To explain the variations in each company’s monthly return (r_t), we use a multi-factor general model shown in equation (1):

In this model, β measures an asset’s sensitivity to systematic risk, which is associated with the overall market and proxied by the ASX200’s monthly return (R_t). The variable D_it represents a binary variable set to 1 for the corresponding i^th calendar month (i = 1,…,12) and 0 otherwise, while α_i denotes the estimated coefficients that capture calendar (seasonal) anomalies in each specific month, where 1 = January, 2 = February,… The term ΔC_t reflects the monthly changes in the RBA’s cash rate, with γ_j measuring the effects of contemporaneous (j = 0) and lagged (j = 1,2,…,k) changes in the cash rate on the companies’ monthly returns. Similarly, O_t represents the monthly logarithmic changes in crude oil prices, and η_j captures the effects of contemporaneous (j = 0) and lagged (j = 1,2,…,k) logarithmic changes in crude oil prices on the returns of a given company. The term λ_j accounts for the effects of the lagged dependent variable (j = 1,2,…,k), and ε_t represents the stochastic residuals.

Similar multifactor models have been used in the literature (see, inter alia, Faff and Chan, 1998; Martin and Keown, 1977). First, we incorporate the market return to capture systematic movements in individual company returns; stocks with betas greater than 1.0 are thus considered more volatile than the market, and vice versa. Second, it is crucial to account for cyclical and seasonal variations in the data generation process (e.g. Urquhart and McGroarty, 2014) using month-specific dummy variables. Equation (1) lacks an intercept, ensuring that inclusion of all 12 dummy variables avoids the dummy variable trap. Thus, the estimated α_i coefficients in equation (1) denote the average effect of a given month on monthly returns after controlling for other relevant factors.

Third, this study aims to investigate the dynamic impacts of interest rate changes on equity returns in Australia’s largest companies. To achieve this, we examine the statistical significance of the γ coefficients. A statistically significant γ₀ indicates the effect is instantaneous, while a significant γ₁₂ suggests delayed effects manifesting after 12 months. Fourth, building on prior research (e.g. Faff and Brailsford, 1999; Sadorsky, 2001; Caporale et al., 2022), we include changes in oil prices as potential regressors of equity returns. While oil prices significantly affect energy stocks (e.g. Woodside Energy), their influence on stocks in other sectors like financials, insurance and materials may vary and requires thorough empirical investigation to uncover dynamic responses.

Finally, we incorporated lagged dependent variables into the general model for two main purposes. First, we aimed to gauge the extent of inertia present in the return series. We achieved this by examining the sum of the statistically significant lambda coefficients, which reflect the influence of past returns on current returns, offering insights into the persistence or momentum in the return series. Second, the inclusion of lagged dependent variables also served to address any potential serial correlation in the stochastic residuals. Against this background, equation (1) can also be rewritten as follows:

The general unrestricted model (GUM), equation (2), includes five sets of variables, constructed based on economic theory, empirical evidence or a combination of both. The simplification process involves refining the general model by eliminating non-significant variables through a series of parameter restrictions and diagnostic tests. These include the Wald test, the Lagrange multiplier autocorrelation test (Breusch and Godfrey, 1986), the heteroscedasticity test (Harvey, 1976) and the parameter stability test (Andrews, 1993; Andrews and Ploberger, 1994). At each step, statistical and maximum likelihood tests guide the model selection process. These tests assess the significance and stability of the model, while criteria such as the AIC, SIC and HQC help in balancing between model complexity and goodness of fit. Following the GETS approach (Escribano and Sucarrat, 2011; Hoover and Perez, 1999), we impose valid restrictions that lead to nested, simpler models, thus reaching the goal of parsimonious modeling, wherein we find a model that adequately explains the data while avoiding overfitting.

Following Castle, Doornik, and Hendry (2021), we iteratively remove the most insignificant variable(s), continuously checking that the model still passes the diagnostic tests. If removing a variable causes the model to fail any diagnostic test, we reinstate that variable and exclude it from further elimination in the current path. This process of eliminating variables continues until no insignificant variables remain or until further eliminations cause diagnostic test failures.

Assessing the stability and robustness of the dynamic effects of interest rate changes on equity returns (i.e. γ coefficients) across different subsample periods is essential for ensuring the valid statistical inferences and generalizability of our findings. One commonly used method involves dividing the data set into two or more samples to scrutinize the stability of estimated coefficients. However, arbitrary splitting of the sample period can compromise the reliability of these inferences, raising concerns about the robustness of the estimated relationship. An effective alternative approach to mitigate the arbitrary selection of breakpoints is to use the Quandt–Andrews test (Andrews, 1993; Andrews and Ploberger, 1994), particularly when uncertainty persists regarding the choice of breakpoint(s) beforehand.

The Quandt–Andrews test identifies one or more unknown structural breakpoints within the sample of a specified equation. It operates by conducting individual Chow tests at each observation between two-time intervals generating statistics, which are then aggregated into a single test statistic assessing the null hypothesis of no breakpoints within a trimming range which is typically assumed to be 15%–20%. By default, the test evaluates whether a structural change exists across all or a selected number of parameters of the original equation. Given the focus of our study we run this test on the estimated γ coefficients. The Quandt–Andrews test yields two key statistics: the Likelihood ratio F statistic and the Wald F statistic. The likelihood ratio F statistic compares the sums of squared residuals under restricted and unrestricted conditions. Conversely, the Wald F statistic is derived from a standard Wald test, evaluating whether coefficients remain consistent across subsamples. We use the average of the Quandt–Andrews individual F statistics to test the stability of the γ coefficients as follows:

As τ₁ approaches the start of the sample or τ₂ nears the end of the sample period, the AveF statistics becomes less reliable. To address this issue, it is commonly recommended to exclude the ends of the sample from the testing process. The trimming region is typically set at 15%, entailing the exclusion of the initial and last 15% of observations. Setting the trimming region at 15% ensures a minimum of 60 observations for each sub-sample. If the null hypothesis cannot be rejected say at the 5% level, we can conclude that there is no structural breaks within the sample period.

While alternatives such as time-varying parameter vector autoregression (TVP-VAR) and generalized autoregressive conditional heteroskedasticity (GARCH) models are commonly used to analyze the dynamic nature of stock returns, the GETS approach was selected for its ability to simplify complex models without losing key information. TVP-VAR models address endogeneity and allow for time-varying parameters, while GARCH models focus on volatility clustering. However, both can become overly complex and computationally demanding. In contrast, GETS strikes a balance between complexity and simplicity, ensuring the final model remains both manageable and insightful. This makes GETS particularly well-suited for empirical research where interpretability and robustness are essential.

Average monthly stock prices were sourced from Yahoo Finance. The ASX200 index data was retrieved from Investing.com, daily Brent oil prices were obtained from the Federal Reserve Economic Data (FRED) website (https://fred.stlouisfed.org), and cash rate data was sourced from the RBA website (www.rba.gov.au). Daily data were converted to monthly observations using the average method. We selected the 15 largest Australian companies, provided their stock prices were available as of January 1998. The list of these 15 companies, which constitute about 55% of the ASX200 market capitalization, is presented in Table 1. It also presents the monthly descriptive statistics and unit root test results for all series from January 1998 to May 2024, covering 317 observations. Table 1 includes the mean, standard deviation, skewness, kurtosis, Jarque–Bera statistic and Augmented Dickey–Fuller (ADF) test results for each variable. The stocks ANZ, BHP, CBA, CSL, FMG, MQG, NAB, QBE, RIO, TCL, TLS, WBC, WDS, WES, and WOW have mean returns ranging from 0.0000–0.0250 and standard deviations from 0.0508–0.1908. Most return series are negatively skewed, and exhibit kurtosis well above 3, suggesting heavier tails and a higher probability of extreme values compared to a normal distribution. These characteristics lead to the rejection of the normality assumption at the 5% level or better, based on the Jarque–Bera statistics.

The monthly mean return per unit of standard deviation for CSL is the highest at 0.1863, followed by WOW, FMG, CBA and WES. In contrast, TLS has the lowest ratio, followed by NAB, QBE, WDS and WBC. During the entire sample, the ASX200 index and the cash rate show mean levels of 4939 and 3.65%, respectively, with the Brent oil price averaging $62.6 per barrel. The ADF test indicates that all variables appearing in equation (1) are stationary. In addition, we applied several other tests with and without structural breaks, all of which confirm the stationarity of these variables. These additional test results are available from the authors upon request.

Figure 1 displays the individual time plots of monthly series during the sample period (Jan 1998–May 2024), indicating no missing data. These graphs clearly illustrate the significant volatility clustering during two major economic disruptions: the 2008 global financial crisis (GFC) and the COVID-19 pandemic beginning in 2020. On the left vertical axes of the graphs, the kernel density distributions provide a deeper insight into the underlying data distribution. These distributions once again confirm that the return series are predominantly negatively skewed, with a long tail extending toward the lower end. This skewness suggests that while most values are relatively high, there are some exceptionally low values that pull the mean downward. The estimated high kurtosis statistics are typical of financial turbulence, marked by elevated frequency and magnitude of extreme events, resulting in less predictable and more erratic data patterns.

Tables 2 and 3 present the estimated GUMs for each of the 15 Australian companies. These models include the market beta, all present and lagged values of changes in the RBA’s cash rate and the price of oil, and dummy variables capturing calendar anomalies, regardless of their statistical significance. The estimated β coefficients are all significant at the 1% level with the correct sign and magnitude. As expected, defensive stocks such as WOW and TLS have the lowest betas, whereas MQG and FMG have the highest betas, a result that remains consistent even in the parsimonious specific model. We then applied the general-to-specific modeling strategy to estimate the specific models, retaining only statistically significant variables. The specific models should also outperform the GUMS based on three model selection criteria (i.e. the SIC, AIC and HQC) and pass residual-based diagnostic tests, including Breusch and Godfrey’s (1986) serial correlation Lagrange multiplier (LM) test, Harvey’s (1976) heteroskedasticity test, and the Quandt–Andrews stability test (Andrews, 1993; Andrews and Ploberger, 1994).

Tables 4 and 5 present the final estimation results for the specific models. As expected, the estimated β is consistently significant across most companies. Although the calendar anomalies are stock-specific, the results indicate that α₈, α₁₀ and α₁₁ are statistically significant for most stocks. This suggests that stock returns are generally higher in August, October and November. These findings support the notion of an end-of-year rally and muted performance in the months around May, resonating with the famous saying, “Sell in May and go away.”

The calendar effects observed, particularly higher returns in August, October and November, and lower performance around May, can be attributed to a combination of seasonal investor behavior and economic cycles. The “end-of-year rally” (Valadkhani and O'Mahony, 2024) is driven by factors such as institutional portfolio rebalancing, tax-loss harvesting, and positive earnings reports, which boost market sentiment in the latter part of the year. August benefits from mid-year earnings disclosures, fueling optimism. Conversely, the “Sell in May and go away” effect (Bouman and Jacobsen, 2002; Andrade, et al., 2013; Dichtl and Drobetz, 2015) is linked to lower trading volumes during the year, leading to reduced market liquidity and more cautious investor behavior, often resulting in muted performance from May to October. These patterns reflect the influence of investor psychology and market cycles on stock returns.

The estimated sum of γ coefficients in Tables 4 and 5 reveals the following important findings in relation to the dynamic impacts of changes in the cash rate on the equity return of the sample Australian companies: First, in order of magnitude the following stocks demonstrate negative cumulative impacts (the sum of γ coefficients) in response to changes in the cash rate: MQG (−0.082), WES (−0.043), CBA (−0.042), WOW (−0.012) and BHP (−0.011). Since MQG is an investment bank and WES is classified in the consumer discretionary sector, it is not surprising that both MQG and WES to be the most interest rate sensitive. The estimated effects can manifest in the short term or over mixed short and long terms depending on j in γ_j, highlighting challenges in maintaining profitability or investor confidence amidst rising interest rates in these stocks.

Second, WBC, CSL, RIO and WDS show no significant responses to changes in the cash rate, indicating that their returns are minimally influenced by fluctuations in the RBA’s cash rate. This is likely attributed to their dependence on other market factors or global dynamics that outweigh the impact of domestic interest rate changes. We present below some possible explanations for these findings. Woodside Energy operates within the energy sector, where its revenue and profitability are primarily driven by factors such as oil and gas prices, project cycles, contract acquisitions and operational performance, rather than being significantly impacted by changes in domestic interest rates. WBC’s stock returns are influenced by broader economic factors and global financial conditions rather than changes in the RBA’s cash rate alone. Its diversified business model across various financial services helps mitigate the direct impact of domestic interest rate changes. CSL’s revenue and profitability depend on global demand for its specialized healthcare products, driven by healthcare trends, research outcomes, regulatory approvals and global market dynamics in pharmaceuticals. Similarly, RIO’s stock returns are heavily influenced by global commodity prices (e.g. iron ore and copper), supply and demand dynamics, geopolitical factors and operational efficiencies in its diversified mining operations.

Third, a group of stocks, including QBE (0.0497), TLS (0.0322), NAB (0.0184), FMG (0.0123), ANZ (0.0058) and TCL (0.0022), demonstrate positive impacts in response to changes in the cash rate (the sum of γ coefficients in parentheses). One possible explanation relates to the fact that NAB and ANZ see gains from wider net interest margins when interest rates rise. Higher cash rates may allow these banks to charge higher interest rates on loans while potentially offering lower rates on deposits, thereby widening their interest rate spreads. This leads to increased profitability from lending activities as they earn more from their lending operations. These wider margins can contribute positively to their stock performance, reflecting improved financial performance in a higher interest rate environment. QBE, as an insurance company, can also gain from higher cash rates by investing premiums in fixed-income securities. When interest rates rise, these investments yield higher returns, thereby boosting QBE’s investment income and profitability. FMG’s fortunes, as a pure iron ore exporter, mainly depend more on the Chinese economy than on the cash rate in Australia. Finally, Transurban (TCL), which operates toll roads, benefits from increased economic activity and higher consumer spending during stronger economic conditions marked by higher cash rates.

In addition, the results in Tables 4 and 5 reveal that rising oil prices can exert positive impacts on returns in the following eight companies (the sum of the η coefficients in parentheses): FMG (0.2539), WDS (0.1910), MQG (0.1392), BHP (0.1225), QBE (0.1212), WOW (0.0208), NAB (0.0093) and RIO (0.0081). These positive impacts are likely to evolve over time due to the significant role these companies play in sectors that benefit from higher oil prices, such as mining, energy and financial services with investment portfolios linked to commodities. It is important to note that prolonged declines in oil prices can coincide with an elevated likelihood of a recession. Therefore, rising oil prices not only boost the profitability of oil and commodity-related sectors but also signal stronger economic activity, which supports broader market sentiment and enhances the returns of these companies.

On the other hand, our analysis of the results in Tables 4 and 5 indicates that banks are generally immune from the negative impacts of changes in oil prices, as evidenced by the sum of eta coefficients being mostly zero. This suggests that banking sector returns are more influenced by factors other than oil prices, such as interest rate changes, loan demand and broader economic conditions. In contrast, the following four stocks are negatively influenced by rising oil prices, in order of magnitude: TCL (−0.0624), TLS (−0.0623), WES (−0.0242), and CSL (−0.0143). These negative impacts can be attributed to increased operational costs and reduced consumer spending power due to higher oil prices, which adversely affect transportation, telecommunications, retail and healthcare sectors. When oil prices and inflation rise, households typically cut spending on consumer discretionary items before consumer staples. This behavior explains the negative effects on WES, a retailer more dependent on discretionary spending, and the positive effects on WOW, which deals in consumer staples.

A quick glance at the estimation results in Tables 4 and 5 indicates that only the lagged values of changes in the cash rate are statistically significant in the specified models, except for BHP, where the impact is instantaneous and γ₀ is non-zero. This suggests that we are dealing with a recursive system, making the simultaneity problem inapplicable and, hence, the OLS method both efficient and unbiased (Pindyck and Rubinfeld, 1998). We also observed that η₀ is statistically significant only for WDS, RIO and WOW; for all other stocks, the impacts are associated with lags. Since the price of oil is determined globally and is unlikely to be influenced by returns from these three relatively minor players, the exogeneity of changes in the oil price is quite plausible.

The results in Table 6 show that the estimated specific models consistently outperform the unrestricted general model, regardless of whether we compare AIC, SIC or HQC pairwise. Residual based diagnostic test results in Table 6 further confirm that the parsimonious specific models exhibit no issues with serial correlation or heteroskedasticity. Moreover, according to Table 6, the specific models shown in Tables 4 and 5 successfully pass nearly all the Quandt–Andrews stability tests, which uses an F statistic to test the null hypothesis of parameter stability.

The analysis of sectoral relationships and spillover effects underscores the interconnectedness of industries in response to macroeconomic factors like interest rate changes and oil price fluctuations. For example, shifts in interest rates not only directly impact banks–where rate hikes enhance profit margins–but also influence consumer discretionary sectors such as retail, which may suffer due to reduced spending power. This creates potential spillover effects between financials and discretionary sectors, with rate changes reverberating across the broader economy. Similarly, oil price shocks present a dual impact: energy and mining sectors benefit from higher commodity prices, boosting revenues, while sectors reliant on transportation or consumer spending, such as telecommunications and infrastructure, face rising operational costs. Moreover, seasonal calendar anomalies, like end-of-year rallies, affect different sectors differently, with consumer-facing industries often benefiting, while resource-driven sectors remain more sensitive to global commodity price trends. These interdependencies highlight the broader, cascading effects of macroeconomic shifts, emphasizing the importance of understanding sectoral dynamics and adjusting portfolio strategies accordingly to manage risk and optimize performance in changing economic environments.

This study investigates the dynamic effects of changes in the cash rate and oil prices on the stock returns of Australia’s 15 largest companies. Using a general-to-specific modeling approach, we analyze an updated monthly data set spanning from January 1998 to May 2024. The estimated specific models reveal how each stock responds to variations in the cash rate and oil prices over time, while also controlling for calendar anomalies and cyclicality, as well as the systematic movements in the total ASX market. These multi-factor models follow the principle of parsimony, successfully passing a series of diagnostic tests, including a parameter stability test.

The study identifies significant seasonal trends in stock returns, with August, October, and November showing consistently higher returns, supporting the “Sell in May and go away” strategy and indicating end-of-year rallies. Some stocks demonstrate heightened sensitivity to rising interest rates, particularly in sectors such as consumer discretionary, which are more vulnerable to increased borrowing costs and inflationary pressures. In contrast, other stocks show minimal response to cash rate changes, suggesting that their returns are influenced by global market dynamics or sector-specific conditions. Regarding oil prices, certain sectors benefit from rising oil prices, while others, such as those in telecommunications, retail and healthcare, exhibit negative reactions due to increased operational costs and reduced consumer spending power.

Based on our findings, investors should adjust their portfolios in response to changes in interest rates, oil prices, and specific calendar months to optimize performance in varying economic conditions. When interest rates rise, reallocating from interest-sensitive sectors to those driven by global trends or commodities can help stabilize returns. During periods of rising oil prices, adjusting exposure to sectors with higher operational costs and reduced consumer spending power is crucial. By incorporating these strategies, investors can align their portfolios with economic trends and sector dynamics, enhancing resilience and profitability across different market conditions.

The GETS methodology, while powerful, has several important limitations that can be grouped into five key areas. First, overfitting is a significant concern. The iterative model simplification process may lead to models that fit the sample data well but perform poorly on new, unseen data, thus failing to accurately reflect the true DGP. That is why, to address this issue, this study applies the Quandt–Andrews stability test. Second, the specific final model is highly influenced by the sequence in which variables are removed and the quality of the initial general model specification. Different paths of simplification can lead to varying results, raising concerns about the model’s robustness, placing considerable responsibility on the researcher to ensure an accurate initial specification. Third, GETS assumes linear relationships between variables. While there are extensions to handle non-linearities and structural breaks, these add complexity to the modeling process and can be challenging to implement effectively. Fourth, researchers may use their judgment to decide which variables to retain or discard. Without strong theoretical justification for each variable’s inclusion, this subjectivity may introduce misspecification bias, undermining the model’s reliability. The presence of outliers can also influence the model selection process, and while robust methods can help mitigate this, they introduce further complexity. Fifth, although GETS incorporates techniques for handling structural breaks, such as dummy variables or regime-switching models, accurately identifying and modeling these breaks remains challenging. Despite these limitations, GETS (also known as the LSE method, due to its association with the London School of Economics) remains a valuable tool in econometric modeling, particularly when applied with caution and in conjunction with other techniques to validate results.

The authors thank the editor and two anonymous referees, whose invaluable input and comments considerably improved an earlier version of this article. The usual caveat applies.