From Boom to Bust: Unravelling the Cyclical Nature of Fiji’s Money Demand

Abstract

This study investigates cyclical asymmetries in money demand models considering the moderating effect of financial development. Prior research has overlooked this issue in the money demand literature within the Fijian context, where research is outdated. Using annual data from 1983 to 2023, we find that income elasticity is about positive unity, irrespective of recessions or expansions. In expansions, an increase in interest rates reduces money demand. An increase in interest rates reduces money demand nine times more strongly in recessions. These effects are accentuated with financial development. Declining interest rates do not impact money demand. The findings suggest that stable money demand could be achievable, but only once the impact of structural breaks is accounted for. Under ideal conditions—without such breaks—money demand exhibits stability, and its connection to income and interest rates appears predictable. However, in reality, structural disruptions complicate this relationship, making money demand less consistent with its key drivers and undermining the reliability of money supply as a monetary policy instrument. The findings align with the pulling on a string hypothesis that monetary contractions control inflation, but expansions may not impact output.

1. Introduction

This paper investigates whether money demand’s responsiveness to interest rates is asymmetric over business cycles in Fiji. Understanding these dynamics is crucial for Fiji given its economic volatility history and vulnerability to tourism (Narayan & Narayan, 2008). Speculative demand for money could become highly elastic during recessions, rendering monetary policy ineffective (Woodford, 2022). Additional money supplied by the central bank is hoarded by people, creating a liquidity trap (Hommes & Lustenhouwer, 2019). However, money demand becomes more sensitive to interest rates in expansions. Thus, monetary policy could control inflation, but is ineffective in stimulating economic activity, consistent with the pulling on a string view (Laine & Pihlajamaa, 2024).

This study, therefore, revisits money demand models in Fiji by exploring the asymmetric effects of interest rates across business cycles within the Fijian context. Cyclical asymmetries refer to the fact that money demand does not respond symmetrically across the business cycle. We expect interest rates to have asymmetric effects on money demand in business cycle expansions and contractions due to the diverging motives of holding money in these settings. For instance, firms and households may demand more money for precautionary reasons in recessions, thereby becoming less responsive to interest rates. In contrast, in expansions, rising economic activity may incentivise shifting money to interest-bearing assets, making money demand more sensitive to interest rates.

Moreover, exploring the case of asymmetric effects of interest rates on money demand is important based on the endogenous nature of money (Badarudin et al., 2013). This implies that the quantity of money is determined by the demand for loans endogenously within the economy and is not set outright by central banks (Badarudin et al., 2013). Central banks implicitly influence the creation of money indirectly by setting short-term interest rates (Badarudin et al., 2013). This further implies that during crisis periods, even aggressive monetary easing may fail to stimulate money demand. This underscores the importance of exploring the asymmetric effects of interest rates on money demand.

Prior research has not explored this critical issue within the Fijian context. Narayan and Narayan (2008), for instance, fail to find cointegration between money demand, income, and interest rates. Singh and Kumar (2010) and Rao and Singh (2006), in contrast, find a stable long-run money demand function in Fiji. This could reflect specification issues. Asymmetric effects may raise model specification issues. De Paoli and Zabczyk (2013), for instance, underscore that while linearised models generally replicate the salient features of macroeconomic dynamics, their ability to “match the data” is weaker than nonlinear models. This could lead to biased inference and erroneous conclusions on the stability of money demand models.

We chose Fiji as a case study because it is subject to heightened macroeconomic shocks and economic volatility due to tropical cyclones and political instability (Dornan, 2020). The effects of these may materialise in uncertainty on inflation dynamics and may lead to erratic growth patterns (Dornan, 2020). The situation is further worsened by Fiji’s unhealthy reliance on tourism as a driver of growth (N. N. Kumar & Patel, 2022). For these reasons, stabilising inflation whilst avoiding erratic growth patterns requires an understanding of money demand’s responsiveness to interest rates in business cycle expansions and contractions. Consequently, knowledge of how the critical relationship between money demand and interest rates evolves in alternative phases of the business cycle is important to achieve a stable and low-inflation growth trajectory.

The study mainly contributes by finding that money demand responsiveness to interest rates is asymmetric over the business cycle in Fiji. It uses an updated dataset from 1983 to 2023 and a nonlinear autoregressive distributed lag model and identifies breaks using the Bai-Perron test. A negative relationship between money and interest rates is observed only when interest rates increase in expansions. However, an increase in interest rates reduces money demand about nine times more strongly in recessions compared to expansions. Income elasticity is about positive unity in expansions and recessions, consistent with Rao and Singh (2006). Our study differs from prior studies in the Fijian context by highlighting the asymmetries present in the interest rate and money demand relationship. The results are robust to endogeneity concerns through the ARDL model and to alternative business cycles extraction methods.

The second contribution focuses on the moderating influence of financial development on interest rates. Financial development could moderate the effect of interest rates on money demand due to its impact on the opportunity cost of money by creating near-cash alternatives (Ma & Lin, 2016). Financial development accentuates the negative effect of interest rates on money demand. A more developed financial system enables households to substitute cash for interest-bearing assets. Our results contrast with those of Ma and Lin (2016), who find that financial development weakens the effectiveness of monetary policy. The financial development index follows Sobiech (2019) and is constructed using principal components analysis, incorporating the financial sector size, financial institution depth, and financial institutional efficiency. The study illuminates research on the interplay between financial development and money markets, such as Du et al. (2024), by highlighting its reinforcing effect on money demand interest rate sensitivity. It reaffirms that financial development raises the opportunity cost of holding money.

We build on the theoretical foundations of Keynes, Baumol–Tobin, and Friedman. From the Keynesian framework, we develop our baseline specification comprising income and interest rates as determinants of money demand. We follow the Baumol–Tobin framework by arguing that individuals trade off holding cash, which generates no returns, against interest-bearing assets. We extend our specification following Friedman’s portfolio theory of money, arguing that money is just one of various assets that individuals hold in their portfolios. Aligning with this, we follow S. Kumar et al. (2013) and include exchange rates and the inflation rate in the specification. Overall, our findings are consistent with Neftci (1984), as we find asymmetric effects over the business cycle.

In expansions, the findings align with the “pulling on a string” theory that monetary policy may not be able to stimulate economic activity but may control inflation (Laine & Pihlajamaa, 2024). Our findings indicate that in business cycle expansions, an increase in interest rates reduces money demand, whereas a decrease in interest rates has no effect on money demand. In recessions, an increase in interest rates reduces money demand, but a decrease in interest rates has no effect on money demand. Our findings, therefore, agree with the “pulling on a string” view that monetary policy may control inflation but not promote economic activity (Karras & Stokes, 1999; Laine & Pihlajamaa, 2024).

The findings may be important to policymakers contemplating interest rate and financial development policies. Interest rate sensitivity is contingent on business cycles, and its direction of change is asymmetric. While we reaffirm money demand stability, which implies that monetary aggregates should be the preferred monetary policy tool (Rao & Singh, 2006), Fiji employs exchange rate-based monetary policies. The Reserve Bank of Fiji does not formally target inflation or interest rates, and the interest rate is used as an indicative signal for liquidity operations. Thus, the central bank does not actively use the interest rate as a primary discretionary policy instrument. What our results do imply is that money demand’s sensitivity to interest rates is contingent on the phases of the business cycle, the direction of change of interest rates, and the extent of financial development. This may be used by policymakers to develop broad-based policies aimed at economic stabilisation, for instance, in dealing with external shocks that impact aggregate spending.

In what follows, Section 2 presents the literature review and hypotheses, Section 3 presents the research design, Section 4 presents the data and results, and Section 5 concludes with policy implications.

2. Literature Review and Hypotheses Development

Keynes (1936) developed the liquidity preference theory, which explicitly highlights the role of the transactionary, speculative, and precautionary motives of holding money. Another seminal study by Laidler (1977) highlighted that Keynes did not consider the transactionary and precautionary motives as technically fixed in their relationship with the level of income, emphasising instead that the most important innovation in Keynes’ work was the speculative demand for money. The main argument for the speculative demand for money is that there is a negative relationship between money demand and the interest rate (S. Kumar et al., 2013).

However, the contrasting perspective offered by monetary theorists such as Friedman (1956) argued that money does not impact real economic activity and presented the quantity theory as an alternative theory of money demand. Monetary theorists model money as abstract purchasing power, meaning people hold it with the intention of upcoming purchases of goods and services (S. Kumar et al., 2013). According to S. Kumar et al. (2013), this nests the quantity theory of money demand within the context of neoclassical consumer and producer behaviour in microeconomic theory. Friedman (1956) argued that the velocity of money is highly stable and insensitive to interest rates. This implies that the quantity of money can be modelled accurately by a money demand function (S. Kumar et al., 2013).

S. Kumar et al. (2013) argue that a stable money demand function implies that the appropriate monetary policy tool is the quantity of money as opposed to interest or exchange rates, as this minimises economic fluctuations. Prior research in Fiji is, however, mixed. Singh and Kumar (2010) and Rao and Singh (2006) report that the money demand function in Fiji is relatively stable, and they conclude that the Reserve Bank of Fiji (RBF) should target monetary aggregates. On the other hand, Narayan and Narayan (2008) report that Fiji does not have a stable long-run money demand function, attributing this to the inherent instability of the Fijian economy.

More recently, Movaghari et al. (2024) revisited the empirical relationship between interest rates and money demand and found a hump-shaped link between interest rates and the M1 monetary aggregate in the USA. They note that various structural shifts, such as the removal of Regulation Q and the introduction of sweep technology, potentially blurred the relationship between interest rates and money demand. Salas (2025) finds that precautionary demand for money increases with liquidity needs in household portfolios comprised of cash, government bonds, and equities. Chen and Valcarcel (2025), however, report that money demand, measured with the Divisa monetary aggregates for the USA, is stable, implying that money demand instability may arise due to the aggregation of monetary aggregates rather than a structural shift in economic agents’ preference for monetary assets.

Overall, while the stability of money demand appears to be a contentious issue, the role of asymmetric effects within the money demand–interest rate relationship within the Fijian context has not currently been explored. Woodford (2022) argues that speculative demand for money could become highly elastic in recessions, as any additional money supplied by central banks is hoarded. However, money demand becomes highly sensitive to interest rates in expansions (Hommes & Lustenhouwer, 2019). Thus, monetary policy could control inflation but may not stimulate economic activity (Laine & Pihlajamaa, 2024). Consequently, the first hypothesis of this study is stated below in the null form:

H1. 

The effect of interest rates on money demand is the same in expansions and recessions.

Extending the argument above, another pertinent issue is the direction of change of interest rates in expansions and in recessions (c.f. Zakir & Malik, 2013). Business cycle expansions describe situations of increased business opportunity and confidence. As such, it is possible that this more strongly raises the opportunity cost of holding money. In contrast, a decrease in interest rates may not necessarily reduce money demand as strongly because they may have higher confidence and steady cash flows, implying a lower sensitivity to interest rates. During economic downturns, in contrast, people and firms may be more cautious about holding idle money, thereby raising their sensitivity to interest rates. Given the conflicting theoretical arguments, Hypothesis 2 is stated in the null form below:

H2. 

The effect of an increase in interest rates on money demand is quantitatively the same as a decrease in interest rates in expansions and recessions.

Another pertinent issue is the role of financial innovation and development. Ma and Lin (2016), for instance, report that the effectiveness of monetary policy in influencing output and inflation declines as financial systems become more developed. This reduced sensitivity implies that money demand becomes increasingly decoupled from interest rates. S. Kumar et al. (2013) reaffirms that financial development may blur/weaken the relationship between money demand and interest rates because it creates alternatives to holding money. Firms and households hold less precautionary balances because they can access liquidity more flexibly via near-cash alternatives such as credit lines or overdrafts (Ma & Lin, 2016). In other words, if financial development improves liquidity and reduces the need for idle cash, it weakens the link between money demand and interest rates.

In contrast, Fiador et al. (2022) report that financial development aids in the effectiveness of monetary policy. Financial innovation may bring about an increase in the interest rate elasticity of money demand because it raises the opportunity cost of holding money (Dou, 2018). As financial innovation increases the number of near-cash alternatives, this could conversely raise the opportunity cost of holding money (Fiador et al., 2022). As a result, changes in the interest rate would lead people to quickly shift their portfolios between money and interest-bearing assets. Advanced payment systems and innovations like online banking reduce transaction costs, amplifying this effect.

Financial innovation precedes financial development (Mollaahmetoğlu & Akçalı, 2019). Innovations such as mobile banking, microfinance, fintech platforms, and peer-to-peer lending expand access to credit and savings facilities in underdeveloped or rural areas (Frame & White, 2004). This encourages financial inclusion, a key pillar of financial development (Lenka, 2022). Moreover, financial innovations improve matching between savers and borrowers, such as through improved settlement services, reduced transaction costs, and improved financial intermediation. Moreover, insurance sector innovations allow for improved risk management by firms and households. Innovations such as exchange-traded funds deepen capital markets, promoting capital efficiency (Konstantakopoulou, 2023).

Overall, financial development could moderate the effect of interest rates on money demand because it creates near-cash alternatives. As a result, people could hold less precautionary money balances, which could weaken the link between interest rates and money demand. Conversely, financial development could increase the opportunity cost of holding money, increasing its sensitivity to interest rates. Given these reasons, it is reasonable to assume that financial development may moderate the effect of interest rates on money demand. Hypothesis 3 is therefore stated as:

H3. 

Financial development does not moderate the effect of interest rates on money demand.

3. Methodology

3.1. Models

The study applies an asymmetric cointegrating model. In the first stage, thresholds are determined based on the position of real GDP relative to its long-term trend value. An expansionary regime is noted if real GDP>trend value. A recessionary regime, in contrast, emerges if real GDP<trend. In stage two, the corresponding recessionary and expansionary component of each variable is extracted. The CPI inflation rate and the real exchange rate are additional indicators of the cost of money. All variables are defined in

Table 1. Variable Definitions.

Variable	Definition	Source	Range
$M$	Natural logarithm of real narrow money. Narrow money is measured using M1 sourced from the Asian Development Bank’s key indicators database. The GDP deflator is used with 2015 as a base year.	KIDB & WDI	1983–2023
$y$	Natural logarithm of real GDP measured in constant 2015 Fijian dollars.	WDI	1960–2024
$i$	Bank lending rate. This rate usually meets the short- and medium-term financing needs of the private sector and is normally differentiated according to the creditworthiness of borrowers and objectives of financing.	WDI	1980–2024
$\pi$	Consumer price index (CPI)-based inflation rate measuring annual percentage changes in prices. The Laspeyres formula is used to compute consumer prices.	WDI	1960–2024
$rx$	Real effective exchange rates, which are the nominal effective exchange rate measured against a weighted average of foreign currencies divided by the GDP deflator.	WDI	1980–2024
$f$	Principal component analysis-based financial development index. Financial sector size is measured using the ratio of liquid liabilities to GDP. Financial institutional depth is measured as the ratio of domestic credit granted to private sector to GDP. Financial institution depth is measured by the lending and deposit interest rate spread.	WDI	1980–2024

Note: KIDB—Asian Development Bank’s Key Indicators Database, WDI—World Development Indicators.

below.

The baseline expansionary and recessionary models for Hypothesis 1 are specified below:

$$M_t^b=\eta +{{\vartheta }_1}^b{y}_t^b-{{\vartheta }_2}^b{i}_t^b-{{\vartheta }_3}^b{\pi }_t^b-{{\vartheta }_4}^b{rx}_t^b+u_t$$

(1)

$$M_t^r=\zeta +{{\beta }_1}^r{y}_t^r-{{\beta }_2}^r{i}_t^r-{{\beta }_3}^r{\pi }_t^r-{{\beta }_4}^r{rx}_t^r+u_t$$

(2)

where:

$$x_t^b=x_{t}I\left(y_t>{\mathrm{y}}_t^T\right)$$

(3)

$$x_t^r=x_{t}I\left(y_t<{\mathrm{y}}_t^T\right)$$

(4)

where M is the natural log of real money demand computed as narrow money (M1) divided by the GDP deflator, y is the natural log of real GDP, $i$ is short-term bank loan rates, $\pi$ is the CPI-based inflation rate, $rx$ is the natural log of real effective exchange rate measured in terms of foreign to local currency, x is a placeholder, “b” and “r” denote expansions and recessions, $y_t^T$ is the cyclical part of real GDP extracted with either the Hodrick-Prescott or Christiano–Fitzgerald filter, and I(.) is an indicator variable separating the two regimes. The Ravn–Uhlig approach is used to calculate the lambda parameter for the HP filter.

Directional asymmetries help assess whether positive/negative shocks to GDP or interest rates impact money demand asymmetrically in recessions and expansions.

The expansionary models with positive and negative shocks for Hypothesis 2, respectively, are specified:

$$M_t^{b+}=\eta +{{\vartheta }_1}^{b+}{y}_t^{b+}-{{\vartheta }_2}^{b+}{i}_t^{b+}-{{\vartheta }_3}^{b+}{\pi }_t^{b+}-{{\vartheta }_4}^{b+}{rx}_t^{b+}+u_t$$

(5)

$$M_t^{b-}=\zeta +{{\beta }_1}^{b-}{y}_t^{b-}-{{\beta }_2}^{b-}{i}_t^{b-}-{{\beta }_3}^{b-}{\pi }_t^{b-}-{{\beta }_4}^{b-}{rx}_t^{b-}+u_t$$

(6)

The corresponding recessionary models are:

$$M_t^{r+}=\eta +{{\vartheta }_1}^{r+}{y}_t^{r+}-{{\vartheta }_2}^{r+}{i}_t^{r+}-{{\vartheta }_3}^{r+}{\pi }_t^{r+}-{{\vartheta }_4}^{r+}{rx}_t^{r+}+u_t$$

(7)

$$M_t^{r-}=\zeta +{{\beta }_1}^{r-}{y}_t^{r-}-{{\beta }_2}^{r-}{i}_t^{r-}-{{\beta }_3}^{r-}{\pi }_t^{r-}-{{\beta }_4}^{r-}{rx}_t^{r-}+u_t$$

(8)

where the partial sum decompositions are defined as:

$$x_t^{{b/r}^{+}}={\sum }_{j=1}^t{∆x}_j^{{b/r}^{+}}={\sum }_{j=1}^t\mathrm{m}\mathrm{a}\mathrm{x}({\mathsf{\Delta }x}_j^{{b/r}^{+}},0)$$

(9)

$$x_t^{{b/r}^{-}}={\sum }_{j=1}^t{∆x}_j^{{b/r}^{-}}={\sum }_{j=1}^t\mathrm{m}\mathrm{i}\mathrm{n}({\mathsf{\Delta }x}_j^{{b/r}^{-}},0)$$

(10)

The model for Hypothesis 3 is similar to the models described above except that financial development is considered the moderator within the relationship between interest rates and money demand.

$$M_t^b=\eta +{{\vartheta }_1}^b{y}_t^b-{{\vartheta }_2}^b{i}_t^b-{{\vartheta }_3}^b{i}_t^b\times f_t^b-{{\vartheta }_4}^b{\pi }_t^b-{{\vartheta }_5}^b{rx}_t^b+u_t$$

(11)

$$M_t^r=\eta +{{\vartheta }_1}^r{y}_t^r-{{\vartheta }_2}^r{i}_t^r-{{\vartheta }_3}^r{i}_t^b\times f_t^b-{{\vartheta }_4}^r{\pi }_t^r-{{\vartheta }_5}^r{rx}_t^r+u_t$$

(12)

where financial development accentuates (attenuates) the effect of interest rates on money demand if the interaction effect and standalone effect of interest rates on money demand have the same (opposite) sign. Financial development is further considered the moderator in the expansionary/recessionary models with positive/negative shocks of interest rates. In these models, the corresponding decomposition of financial development is undertaken to ensure consistency with the model being estimated.

3.2. Structural Breaks

Structural breaks are identified using the Bai and Perron (2003) break test. This approach helps identify multiple break points and correlate the breaks with actual events to provide insight into the reasons behind the breaks. Structural breaks are measured using dummy variables set to 1 from the identified break date until one period before the next break, if any.

3.3. ARDL Model

The autoregressive distributed lag (ARDL) model is employed for estimation. The ARDL approach can be applied irrespective of whether the data are stationary at the level or first difference (Aliyev & Eylasov, 2025). It allows cointegration analysis while reducing the small sample bias and introducing a lag structure best suited to the specific dataset (Aliyev & Eylasov, 2025). It also mitigates endogeneity, provided the ARDL model does not suffer from autocorrelation (Aliyev & Eylasov, 2025).

$$\mathsf{\Delta }{z}_t={\alpha }_1+{\alpha }_2{z}_{t-1}+{\alpha }_3{x}_{t-1}+{\sum }_{i=1}^{p-1}{\alpha }_{1i}\mathsf{\Delta }{z}_{t-i}+{\sum }_{i=0}^{q-1}{\alpha }_{2i}\mathsf{\Delta }{x}_{t-i}+u_t$$

(13)

where z and x are timeseries variables and $-1<{\alpha }_2<0$ is the adjustment coefficient.

Cointegration via the bounds test is confirmed by the significance of the lagged level variables in Equation (13) using either an F or a WALD test. Cointegration is confirmed if the resulting F-statistic exceeds its upper critical bound, it is rejected if the F-statistic is less than its lower bound and is inconclusive otherwise. To address endogeneity, the ARDL model requires the absence of auto-correlated residuals. Instances of heteroscedasticity may be addressed by changing the lag structure or employing robust standard errors (Aliyev & Eylasov, 2025).

4. Data and Results

4.1. Data and Descriptive Statistics

A sample from 1983 to 2023 (n = 41) is used for analysis. Real GDP has a positive and significant correlation with real money, while the correlations are negative with interest and exchange rates. The inflation rate is not significantly correlated with real money (

Table 2. Descriptive Statistics and Correlation Matrix.

Variables	$\mathit{M}$	$\mathit{y}$	$\mathit{i}$	$\mathit{\pi }$	$\mathit{r}\mathit{x}$	$\mathit{f}$
Panel A. Descriptive statistics
Mean	21.3831	22.6452	9.6122	5.8663	4.7968	−0.0122
Median	21.2001	22.6620	8.3975	6.5709	4.7611	−0.1682
Maximum	22.6078	23.0966	15.1127	18.8096	5.1906	3.1522
Minimum	20.7398	22.2329	5.67732	−16.4487	4.6051	−1.4288
Std.Dev.	0.5066	0.2570	3.1947	6.0424	0.1667	1.0008
Skewness	0.6860	−0.0047	0.3311	−1.1315	1.3751	0.7908
Kurtosis	2.3527	1.8743	1.6127	6.3840	3.7479	3.6961
Jarque–Bera	3.9320	2.1646	4.0371	28.313 A	13.8769 A	5.1016
	[0.1400]	[0.3388]	[0.1328]	[<0.01]	[<0.01]	[0.0780]
Panel B. Correlation matrix
$y$	0.8997 A	1
	[<0.01]
$i$	−0.8648 A	−0.9675 A	1
	[<0.01]	[<0.01]
$\pi$	−0.2550	−0.1533	0.1596	1
	[0.1076]	[0.3386]	[0.3188]
$rx$	−0.6553 A	−0.7308 A	0.8096 A	0.0566	1
	[<0.01]	[<0.01]	[<0.01]	[0.7248]
$f$	0.9225 A	0.8055 A	−0.7791 A	−0.3448 B	−0.6407 A	1
	[<0.01]	[<0.01]	[<0.01]	[0.03]	[<0.01]

Note: ^{A, B}—significant at 1%, 5%.

).

4.2. Unit Roots

The unit root tests, based on the augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) tests, indicate that all variables are stationary in first differences (

Table 3. Unit Root and Cointegration.

Variables	ADF Level	ADF 1st Diff.	PP Level	PP 1st Diff.
$M_t$	0.4431 (0) [0.9824]	−7.9849 A (0) [<0.01]	2.5014 (16) [0.9671]	−9.2471 A (7) [<0.01]
$y_t$	−0.8340 (0) [0.7984]	−6.1360 A (0) [<0.01]	−0.8344 (1) [0.7329]	−6.1383 A (3) [<0.01]
$i_t$	−1.4132 (0) [0.5662]	−5.2151 A (0) [<0.01]	−1.4232 (5) [0.5614]	−5.1340 A (7) [<0.01]
${\pi }_t$	−5.0592 A (0) [<0.01]	−7.3744 A (1) [<0.01]	−5.0584 A (1) [<0.01]	−12.4870 A (5) [<0.01]
${rx}_t$	−2.1008 (0) [0.2454]	−4.4113 A (0) [<0.01]	−2.0986 (4) [0.2463]	−4.1939 A (6) [<0.01]
$f_t$	1.3668 (0) [0.9986]	−3.1935 B (0) [0.03]	0.9414 (1) [0.9951]	−3.2098 B (1) [0.03]

Note: ^{A, B}—indicates significance at 1%, 5%. Lag length for ADF and bandwidth for PP in (.) and determined by AIC, p value in [.]. Unit root test specification assumes intercept only.

).

4.3. Structural Breaks Results

The results for the expansionary models indicate that 2014 was a significant break (

Table 4. Break Test.

Specification	0 vs. 1	1 vs. 2	2 vs. 3	3 vs. 4	Break Dates
Boom	28.4155 A [8.58]	4.0527 [10.13]			2014
Boom-positive	121.9293 A [8.58]	20.3196 A [10.13]	40.9063 A [11.14]	1.3262 [11.83]	1995, 2005, 2014
Boom-negative	20.4779 A [8.58]	10.6019 A [10.13]	3.6010 [11.14]		1996, 2016
Bust *	38.4697 A [11.47]	15.8429 A [12.95]	0.6710 [14.03]		2000, 2013
Bust-positive *	39.1664 A [11.47]	11.5332 A [12.95]	9.1533 [14.03]		2000, 2013
Bust-negative	158.8025 A [8.58]	72.5775 A [10.13]	33.8195 A [11.14]	6.7003 [11.83]	1990, 2008, 2014

Note: ^A indicates a significant break at 5%. * Assumes trend breaks for break test model stability. Boom refers to the expansionary phase of the business cycle. Bust refers to the recessionary phase of the business cycle. These definitions are applied hereafter.

). This reflects the 2014 Fijian elections, which favoured stability and transparency after 8 years of an interim military rule (Madraiwiwi, 2015). Unsurprisingly, the 2014 break year had a positive and significant effect on money demand, as indicated by the structural breakpoint in our NARDL model. From a financial and economic point of view, this transition likely enhanced investor sentiment and household confidence, increasing integration with global financial markets and more efficient monetary transmission.

Regarding the remaining breaks, the year 1999 reduced money demand, potentially reflecting the effects of tropical cyclone Dani (UNOCHA, 1999). The breaks for 1997 and 2005 are insignificant and excluded. Regarding the breaks in the recessionary models, 2000 and 2013 are significant. The year 2000 reduced money demand, reflecting George Speight’s civilian coup (BBC News, 2024). This reflects that political uncertainty erodes confidence, prompting people to hold less currency. The year 2013 could reflect the adverse effects of severe tropical cyclone Ian, which again reduced money demand.

There is clear evidence of causal logic between breakpoints and our results, as explained above. The 2014 break also coincides with advancements in Fiji’s financial sector (e.g., expanded credit access, fintech adoption), which our moderation tests show amplified interest rate effects. This supports the narrative that financial development and political stability jointly influence money demand dynamics.

4.4. Main Tests

The bounds test indicates cointegration in all specifications (

Table 5. Expansionary Models.

Variables	Boom	Boom-Positive	Boom-Negative
Panel A. Long run
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{b}}$	0.9289 A [0.0816]	0.9476 A [0.0139]	0.9358 A [0.0542]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{b}}$	−0.0691 B [0.0277]	−0.0178 A [0.0035]	−0.0061 [0.0125]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{b}}$	0.0153 [0.0140]	0.0093 [0.0092]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{b}}$	0.1838 [0.4269]	−0.0344 [0.0717]	0.0199 [0.2800]
Break (1999–2004)		−0.0938 A [0.0174]
Break (2014)	0.8571 A [0.3103]	0.3489 A [0.0327]
Intercept	−0.0455 [0.0424]	−0.0958 B [0.0353]
Panel B. Short run
$∆{\mathrm{M}}_{\mathrm{t}-1}^{\mathrm{b}}$		−0.0019 B [0.0009]	0.4455 B [0.1894]
$∆{\mathrm{M}}_{\mathrm{t}-2}^{\mathrm{b}}$			−0.0001 [0.0006]
${∆\mathrm{y}}_{\mathrm{t}}^{\mathrm{b}}$	0.9341 B [0.0392]	0.9353 A [0.0162]	0.9332 A [0.0297]
$∆{\mathrm{y}}_{\mathrm{t}-1}^{\mathrm{b}}$			−0.4597 A [0.1859]
$∆{\mathrm{i}}_{\mathrm{t}}^{\mathrm{b}}$	−0.0331 A [0.0149]	−0.0175 A [0.0033]	−0.0035 [0.0051]
$∆{\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{b}}$	−0.0001 [0.0069]	0.0165 A [0.0023]
$∆{\mathsf{\pi }}_{\mathrm{t}-1}^{\mathrm{b}}$		0.0043 A [0.0015]
$∆{\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{b}}$	0.0881 [0.2069]	0.0053 [0.0811]	0.0145 [0.1478]
$∆{\mathrm{r}\mathrm{x}}_{\mathrm{t}-1}^{\mathrm{b}}$			0.1981 B [0.0783]
Break (1999–2004)		−0.0925 A [0.0175]
Break (2014)	0.4110 A [0.1080]	0.3443 A [0.0316]
${\mathrm{E}\mathrm{C}\mathrm{T}}_{\mathrm{t}-1}$	−0.4796 A [0.0882]	−0.4853 A [0.0931]	−0.3994 A [0.1223]
Panel C. Model diagnostics
Adj. R2	0.9689	0.9926	0.9830
Regression standard error	0.0715	0.0168	0.0266
Bounds statistic	5.5876 A	42.913 A	25.6498 A
Upper bound	5.455	5.455	5.816
Lower bound	3.967	3.967	4.428
Serial Correlation	1.5713 (0.4558)	0.1046 (0.7463)	3.2801 (0.1940)
Heteroscedasticity	12.0070 (0.1509)	5.5432 (0.4991)	15.7483 (0.1507)
Functional Form	2.1967 (0.1257)	0.6878 (0.4147)	2.1686 (0.1550)
Normality	1.8499 (0.4567)	1.9083 (0.4215)	1.7879 (0.4781)
Cusum & Cusumsq	Stable	Stable	Stable

Note: ^{A, B}—significant at 1%, 5%. “[.]”—standard error, “(.)”—p-value. Boom: baseline expansionary model, Boom-Positive: expansionary model with positive shocks, Boom-Negative: expansionary model with negative shocks. Negative inflation shocks are not found in the data during boom periods and are thus excluded in column 3. Bounds statistics—Small sample bounds from Narayan (2005).

). The income elasticity is approximately positive unity, aligning with Singh and Kumar (2010) and Rao and Singh (2006), and is similarly sized in expansions and in recessions. This indicates that money demand fluctuates roughly in proportion to income. This aligns with Milton Friedman’s modern quantity theory of money, also termed the portfolio theory of money demand, where income elasticities are approximately positive unity.

The interest rate responsiveness is, however, only significant when interest rates increase in expansions. In other words, the results indicate that monetary policymakers may be unable to fuel further growth during expansions but could limit overheating and control inflation. An increase in interest rates reduces money demand more strongly in recessions than in expansions (

Table 6. Recessionary Models.

Variables	Bust	Bust-Positive	Bust-Negative
Panel A. Long run
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{r}}$	1.0237 A [0.1236]	0.9227 A [0.1392]	0.9563 A [0.0123]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0623 [0.0470]	−0.1672 A [0.0449]	0.0101 [0.0059]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{r}}$	−0.0163 [0.0106]	−0.0117 [0.0163]	−0.0001 [0.0013]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{r}}$	−0.1936 [0.6504]	0.4918 [0.7248]	−0.1160 [0.0677]
Break (2000–2012)	−0.1825 [0.1296]	−1.0732 A [0.1969]
Break (2013)	−0.1825 [0.1336]	−0.2991 [0.2854]
Break (1990, 2007)			−0.2447 A [0.0372]
Break (2008, 2013)			−0.5241 A [0.0428]
Break (2014)			−1.2857 A [0.0446]
Intercept	−0.0478 [0.1047]	0.8313 [0.4311]	−0.0109 [0.0255]
Panel B. Short run
$∆{\mathrm{y}}_{\mathrm{t}}^{\mathrm{r}}$	0.9950 A [0.0830]	1.1301 A [0.0919]	0.9556 A [0.0122]
$∆{\mathrm{y}}_{\mathrm{t}-1}^{\mathrm{r}}$		0.2057 A [0.0721]
$∆{\mathrm{i}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0430 [0.0319]	−0.0841 A [0.0269]	0.0101 [0.0059]
$∆{\mathrm{i}}_{\mathrm{t}-1}^{\mathrm{r}}$		0.0516 B [0.0254]
$∆{\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{r}}$	−0.0112 [0.0068]	0.0103 C [0.0055]	−0.0009 [0.0013]
$∆{\mathsf{\pi }}_{\mathrm{t}-1}^{\mathrm{r}}$		0.0121 B [0.0059]
$∆{\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{r}}$	−0.1336 [0.4424]	−0.6897 [0.4772]	−0.1159 C [0.0676]
$∆{\mathrm{r}\mathrm{x}}_{\mathrm{t}-1}^{\mathrm{r}}$		−1.0797 A [0.3792]
Break (2000–2012)	−0.1260 [0.0823]	−0.5089 A [0.0997]
Break (2013)	0.0959 [0.0998]	−0.1418 [0.1334]
Break (1990, 2007)			−0.2445 A [0.0372]
Break (2008, 2013)			−0.5237 B [0.0428]
Break (2014)			−1.2848 A [0.0445]
${\mathrm{E}\mathrm{C}\mathrm{T}}_{\mathrm{t}-1}$	−0.6901 A [0.1801]	−0.4742 A [0.1059]	−0.8992 A [0.0006]
Panel C. Model diagnostics
Adj. R2	0.9589	0.9908	0.9830
Regression standard error	0.1736	0.1115	0.0373
Bounds statistic	36.7855 A	49.2555 A	36.78550 A
Upper bound	5.455	5.455	5.455
Lower bound	3.967	3.967	3.967
Serial correlation	2.5194 (0.2837)	2.0725 (0.3548)	3.0645 (0.2161)
Heteroscedasticity	8.7099 (0.3674)	6.6505 (0.9667)	6.2720 (0.6168)
Functional Form	2.7543 (0.1590)	0.4438 (0.5125)	2.4696 (0.2650)
Normality	0.0096 (0.9951)	0.0721 (0.9646)	0.9698 (0.6158)
Cusum & Cusumsq	Stable	Stable	Stable

Note: ^A–C—significant at 1%, 5%, 10%. “[.]”—standard error, “(.)”—p-value. Bust: baseline recessionary model, Bust-Positive: recessionary model with positive shocks, Bust-Negative: recessionary model with negative shocks. Bounds statistics-small sample bounds from Narayan (2005).

). During downturns, households may face tighter credit constraints (Gertler & Gilchrist, 2018). A rise in interest rates may make borrowing more expensive, forcing consumers to cut spending more aggressively than in expansions. Negative interest rate shocks do not reduce money demand in either phase of the business cycle. The results agree with the “pulling on a string” view that monetary contractions limit price growth, but monetary expansions may not assist in recovery (Laine & Pihlajamaa, 2024).

The asymmetric effects of interest rates align with Neftci (1984), in that macroeconomic variables are associated with asymmetries in alternative phases of the business cycle. The findings further align with related research, such as Lu et al. (2024) that the association between economic indicators strengthens in recessions where negative spirals amplify co-movements in job losses, decreased spending, and more cautious behaviour amongst economic agents. This seems to align with loss aversion theory, which argues that economic agents are more averse to loss-generating situations such as economic recessions.

The adjustment coefficient (ECT) is similarly sized in expansions. Positive shocks lead to a disequilibrium period of about 2.06 years in expansions. Negative shocks to income lead to a disequilibrium period of 2.51 years in expansions. In recessions, positive shocks lead to a disequilibrium period of about 2.11 years. Negative shocks are corrected in about 1.11 years in recessions. In recessions, households become more risk-averse and prioritise holding liquidity rather than expenditure, accelerating the adjustment of money.

Nonetheless, the post-Keynesian theory of endogenous money supply emphasises the importance of bank loans in impacting money supply changes (Badarudin et al., 2013). Monetary endogeneity implies that the money supply is not exogenously fixed by central banks, but rather, is created within the economy through the bank lending channel. Specifically, the emission (creation) of money occurs when banks issue loans. In doing so, they simultaneously create matching deposits (Badarudin et al., 2013). This process implies that money supply is determined by the private sector’s demand for loans. Because money is endogenously created, central banks cannot precisely control the quantity of money through monetary aggregates (Badarudin et al., 2013). Instead, they influence the creation of money indirectly by setting short-term interest rates (Badarudin et al., 2013).

A further consequence is that endogenous money implies that during crisis periods, even aggressive monetary easing may fail to boost lending because banks cannot force borrowers to take loans (Badarudin et al., 2013). This aligns with the “pulling on a string” view that monetary policy may not help in getting an economy out of a recession, but may help control inflation. Our results indicate that an increase in interest rates in expansions reduces money demand, whereas a decline in interest rates does not impact money demand in recessions. Therefore, monetary easing in crisis periods does not increase money demand, consistent with the endogenous nature of money (Badarudin et al., 2013).

Table 7. Robustness Test—CF Filter Results.

Variables	Boom	Boom-Positive	Boom-Negative
Panel A. expansions
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{b}}$	1.0146 A [0.0863]	0.9476 A [0.0689]	0.8907 A [0.3398]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{b}}$	−0.0664 B [0.0298]	−0.0334 B [0.0113]	−0.0898 [0.1158]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{b}}$	−0.0101 [0.0133]	−0.0173 [0.0125]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{b}}$	−0.2201 [0.4520]	−0.0117 [0.3595]	−0.4509 [0.8365]
Break (1999–2004)		−0.1924 B [0.0869]
Break (2014)	0.4114 A [0.1312]	0.5432 A [0.0916]
Intercept	−0.0159 [0.0482]	−0.0098 [0.1213]	0.0043 [0.7763]
Panel B. recessions
Variables	Bust	Bust−Positive	Bust−Negative
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{r}}$	0.9785 A [0.1236]	0.9357 A [0.0914]	0.7692 A [0.1006]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0669 [0.0443]	−0.0731 A [0.0256]	−0.0612 [0.0386]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{r}}$	−0.0411 B [0.0169]	−0.0032 [0.0088]	−0.0091 [0.0093]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0079 [0.6099]	0.2019 [0.4762]	−0.8997 [0.5396]
Break (2000–2012)	−0.1407 [0.1069]	−0.8133 A [0.1357]
Break (2013)	−0.2216 [0.1355]	−0.7474 A [0.2286]
Break (1990, 2007)			−0.2758 [0.2526]
Break (2008, 2013)			−0.7252 B [0.3020]
Break (2014)			−0.8706 A [0.3049]
Intercept	0.0723 [0.1043]	0.0858 [0.1991]	−0.0729 [0.1916]

Note: ^{A, B}—significant at 1%, 5%. “[.]”—standard error. The adjustment coefficient is appropriately signed and significant in all specifications considered.

indicates that the results are robust to the Christiano–Fitzgerald (CF) filter since the HP filter could bias the extraction of cycles at data endpoints (N. N. Kumar et al., 2024). Notably, Figure 1 indicates that both filters produce similar decompositions of business cycles. The HP filter, however, produces a stronger recessionary outcome in the 2020–2021 period and indicates that 2013 may have been a recessionary period (Figure 1). This aligns with the break test results, where the year 2013 had a significant negative effect on money demand.

Further support for the validity of our results derives from the evolution of the velocity of circulation in recessionary and expansionary periods. We find that the average value of the velocity of circulation is lower in recessionary periods for both the HP and CF filters, reaffirming that consumers and businesses cut back on spending (

Table 8. Average Velocity of Circulation.

Filter	Expansion	Recession
HP	3.6739	3.6498
CF	3.7288	3.6128

).

Financial development may moderate the effect of interest rates on money demand. A more developed financial system indicates that households access various financial instruments that substitute for cash. This could raise the opportunity cost of holding money.

The financial development index encompasses financial sector size, financial institution depth, and institutional efficiency (Sobiech, 2019). Consistent with Sobiech (2019), the ratio of liquid liabilities to GDP measures financial sector size, the ratio of domestic credit to private sector measures financial institutional depth, and the lending-deposit rate spread measures financial institutional efficiency. The index is built using principal component analysis (PCA). The eigenvalue of domestic credit to the private sector is 2.0825, which exceeds positive unity. Therefore, domestic credit to the private sector (financial institution depth) is key to explaining financial development in Fiji based on the Kaiser criterion.

Financial development accentuates the negative effect of interest rates on money demand both in expansions and in recessions (

Table 9. Moderation Test.

Variables	Boom	Boom-Positive	Boom-Negative
Panel A. expansions
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{b}}$	1.0769 A [0.0397]	0.9373 A [0.0194]	0.9109 A [0.0604]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{b}}$	−0.0220 B [0.0087]	−0.0212 A [0.0042]	−0.0020 [0.0149]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{b}}\times {\mathrm{f}}_{\mathrm{t}}^{\mathrm{b}}$	−0.0665 A [0.0077]	−0.0100 A [0.0009]	−0.0021 [0.0027]
${\mathrm{f}}_{\mathrm{t}}^{\mathrm{b}}$	−0.7317 A [0.0608]	−0.0193 [0.0180]	−0.0983 [0.1862]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{b}}$	−0.0115 [0.0377]	−0.0110 [0.0302]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{b}}$	−0.6375 A [0.1998]	−0.0185 [0.0959]	−0.0954 [0.2857]
Break (1999–2004)		−0.1013 A [0.0179]
Break (2014)	0.0067 [0.0117]	0.3340 A [0.0319]
Intercept	−0.0159 [0.0482]	−0.0599 [0.0367]	0.1989 [0.3509]
Panel B. recessions
Variables	Bust	Bust-Positive	Bust-Negative
${\mathrm{y}}_{\mathrm{t}}^{\mathrm{r}}$	1.0586 A [0.0433]	0.8873 A [0.1330]	0.9189 A [0.0187]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0334 A [0.0134]	−0.0826 A [0.0227]	−0.0040 [0.0057]
${\mathrm{i}}_{\mathrm{t}}^{\mathrm{r}}\times {\mathrm{f}}_{\mathrm{t}}^{\mathrm{r}}$	−0.0421 A [0.0107]	−0.0087 A [0.0013]	−0.0011 A [0.0004]
${\mathrm{f}}_{\mathrm{t}}^{\mathrm{r}}$	−0.5490 A [0.0939]	−0.5473 A [0.1256]	−0.0976 A [0.0343]
${\mathsf{\pi }}_{\mathrm{t}}^{\mathrm{r}}$	−0.0151 A [0.0049]	−0.0136 A [0.0043]	−0.0026 [0.0017]
${\mathrm{r}\mathrm{x}}_{\mathrm{t}}^{\mathrm{r}}$	−0.5402 A [0.2243]	0.1114 [0.6613]	−0.0719 [0.0950]
Break (2000–2012)	−0.0231 [0.0307]	−0.4132 A [0.0900]
Break (2013)	−0.0798 B [0.0349]	−0.2261 B [0.1056]
Break (1990, 2007)			−0.2436 A [0.0349]
Break (2008, 2013)			−0.5000 A [0.0388]
Break (2014)			−1.2759 A [0.0429]
Intercept	0.0465 [0.0298]	1.5734 A [0.5296]	−0.1319 B [0.0499]

Note: ^{A, B}—significant at 1%, 5%. “[.]”—standard error. The adjustment coefficient is appropriately signed and significant in all specifications. The eigenvalue of domestic credit to the private sector (percent of GDP) is 2.0825, broad money (percent of GDP) is 0.8916, and the interest rate spread is 0.0259.

). Financial development suggests that people have better access to more attractive alternatives to holding cash, such as bonds/money market instruments with competitive rates. When interest rates rise, people quickly move their money into higher-yielding assets, leading to a stronger decline in money demand.

5. Conclusions, Implications, and Future Research

The findings indicate that the interest rate responsiveness of money demand depends on the interest rate’s direction of change and the phase of business cycles. An increase in interest rates reduces money demand nine times more strongly in recessions than in expansions. The findings imply that the Reserve Bank of Fiji may control inflation with other goals, such as economic stabilisation, being less prominent (Weiping, 2024). Moderation tests suggest that financial development strengthens the negative effect of interest rates. A well-developed financial system makes controlling inflation easier by influencing the responsiveness of money demand to interest rates. The study contributes to money demand research by highlighting asymmetric effects of interest rates over different phases of the business cycle.

The results imply that money demand stability may be feasible, but only after removing the effects of structural breaks. In a world without breaks, money demand remains stable, and its relationship with income and interest rates is argued to be predictable. However, in practice, structural breaks occur that obscure the relationship between money demand and its determinants, rendering money supply a less reliable tool for monetary policy. This aligns with Narayan and Narayan (2008) on the inherent instability within the Fijian economy.

The post-election period saw efforts to stabilise financial markets and improve fiscal discipline, which could have reduced uncertainty and increased the predictability of monetary policy. This aligns with our finding of heightened interest rate sensitivity in expansions post-2014, as financial stability may have encouraged households to respond more actively to interest rate changes.

The policy implications are straightforward. Based purely on the findings, Fiji’s monetary authority can raise interest rates to control inflation and spending in expansionary periods. However, they should avoid raising interest rates in recessions as this is likely to lead to a deeper and prolonged recession. The findings further imply that the appropriate policy mix would be fiscal expansions to get the economy out of a recession, while monetary policy may help control inflation in expansions

Nonetheless, one limitation of this study is that the business cycle series is estimated. While appropriate attention is paid to ensuring the robustness of cyclical decomposition by using the HP and CF filters and applying the Ravn–Uhlig method, which is applicable with annual data, future research could revisit the analysis with higher frequency data. Despite this, the findings underscore the importance of business cycles and financial development with specific reference to Fiji, which has not previously been examined.

Future research could further apply the methodology described in this paper to explore asymmetric effects and nonlinearities within the relationship between money demand and interest rates in other Pacific Island countries, where empirical evidence is scant. Our results demonstrate that the relationship between interest rates and money demand does not hold unconditionally, but rather depends on the state of the economy, whether in recession or expansion, and the direction of change of interest rates. Whether these findings are valid in other developing Pacific island countries is unclear.