Keynes’ Finance Circuit model

Although Keynes (Reference Keynes1936, Reference Keynes1937) seemed to have taken for granted the financing of investment in his General Theory, he pointed out that the entrepreneur may obtain short-run bank loans to finance the net increment expenditure (Keynes Reference Keynes1973, 217). Commercial banks create money every time they increase the aggregate volume of bank loans outstanding in the economy. In the short run, no real resources are involved: no savings in particular take place or have any role in this operation as indicated in the Finance Circuit illustration below. Finance precedes the actual purchase or even the actual production of the investment goods that will be demanded. Investment good producers observe bank credit creation to form their short-run expectations so that more production will only continue after finance has been obtained. Since savings are an allocation of current income, savings and income will be generated only when production has started. Keynes emphasised the role of finance in investment decisions and even provided a definition of finance:

It should be noted that according to Keynes, the Finance Circuit does not begin with money creation but rather commences with the expansion of economic activity which stimulates demand for bank loans. Keynes stressed that demand for finance is associated with activity levels expansion rather than investment (Keynes Reference Keynes1973, 233). Consequently, investment finance is ‘only a special case of the finance required by any productive process’ (Keynes Reference Keynes1973, 208). Keynes likely regards production decisions either of capital or of consumption goods as taken in advance based on orders or of expected demand for bank loans. Only if investment or production are increasing, ‘extra finance involved will constitute an additional demand for money’ (Keynes Reference Keynes1973, 209).

The second point asserted by Keynes is the ability of the entrepreneur to fund the short-run obligations by long-run issues. This point suggests that if money creation is the effective condition for starting the process of investment, it is not all that is required to sustain it. Unlike Kalecki’s model (Reference Kalecki1935) where funding is derived from short-run bank loans and not via long-run bonds and securities, according to Keynes, the firm will be expected to pay back the bank loan by subsequent long-run issues. In this case, the bank will have the liquidity of its balance sheet restored and the firm will have assets and liabilities of comparable maturity.

Keynes, therefore, developed an original sequential process, in which one starts with economic expansion which leads to banks creating finance, and the investment being made creating equivalent savings. The association between investment and savings in Keynes’ prior investment approach is explained by the expenditure multiplier. Keynes (Reference Keynes1936, 115) asserted that, ‘Let us call k the investment multiplier. It tells us that, when there is an increment of aggregate investment, income will increase by an amount which is k times the increment of investment’. An increase in investment, therefore, raises income through the multiplier until an equivalent amount of savings is generated. Investment can be constrained through shortage of credit rather than a shortage of savings (Keynes Reference Keynes1937, 222).

Keynes Finance Circuit can, therefore, be illustrated as follows:

Bank loans → Investment → Economic Growth → Savings → Investment (→ stands for a causal relationship).

In the next subsection, we demonstrate that Keynes Finance Circuit has been adopted and consolidated in post-Keynesian paradigms in which the central thesis is that money supply is endogenously determined by the demand for credit as the economy expands.

Post-Keynesian economists on Keynes’ Financial Circuit model

According to the Post-Keynesian Economics Society (2024), post-Keynesian economics is a school of economic thought that has its origins in models linked to John Maynard Keynes and Michal Kalecki (see also King Reference King2013). For Bougrine and Rochon (Reference Bougrine and Rochon2020), post-Keynesian economics is a theoretical framework that draws its inspiration from the contributions of John Maynard Keynes, as well as other prominent economists such as Michal Kalecki, Joan Robinson, Nicholas Kaldor, and various Cambridge economists. This school of thought emphasises the significance of money in comprehending economic activity. It is therefore not surprising that the post-Keynesian school has remained closest to the spirit of Keynes by adopting and consolidating Keynes’ Finance Circuit. More importantly, banks play a central role in post-Keynesian paradigms related to investment approaches which advocate that entrepreneurs need banking services such as credit which will stimulate investment and boost economic growth as indicated in Keynes Finance Circuit. However, endogenous money supply is a major component of post-Keynesian macroeconomics. It is a tradition that began in the post-Keynesian period with Joan Robinson (Reference Robinson1956) and continued with Kaldor (Reference Kaldor1970), Moore (Reference Moore1988), and Lavoie (Reference Lavoie1992). Today, it has deep and complex roots in post-Keynesian economics and constitutes a rallying point against many dominant schools of economic thought, Rochon (Reference Rochon2023).

The post-Keynesian endogenous theory holds that money supply is endogenously determined by the demand for credit. That is, the existence of money in an economy is driven by the requirements of the real economy. Therefore, money supply is credit-driven and demand-determined. The historical roots of endogenous money after Keynes can be traced back to Joan Robinson, who, in The Accumulation of Capital (1956), advocated a framework not unlike that of modern post-Keynesians and proponents of Keynes Finance Circuit. Following the work of Keynes and Rosa Luxemburg as indicated in Rochon (Reference Rochon and Gibson2005), Robinson placed commercial banks and bank credit at the heart of her views on production and capital accumulation. She supports that as the economy grows, banks supply more credit to meet the growing investment demand of the system. At the very instant that a firm receives credit, money is created by the bank carrying out the payment which will generate investment and economic growth as indicated in Keynes Finance Circuit.

Post-Keynesians further demonstrated their alignment with Keynes’ Finance Circuit when Robinson (Reference Robinson1952) provided the impetus for the demand-following hypothesis expressing Finance Circuit causalities by stating that ‘where enterprise leads finance follows’ (Robinson Reference Robinson1952, 86). Robinson is pointing out that financial development does not support economic growth but rather responds to financial demand for bank services as the economy continues to grow. The demand-following hypothesis suggests that financial deepening occurs due to bank credit demand stimulated by the expansion of the economy, which will then encourage investment and motivate growth as illustrated in Keynes’ Finance Circuit.

Empirical review on Keynes’ Finance Circuit model

The main purpose of the present section is to undertake a critical survey of the empirical literature on a complete Keynes’ finance transmission mechanism influencing policies pursued in Africa in order to highlight potential issues deemed to break new ground in this paper. Having reviewed Keynesian and post-Keynesian models on banks and their extensions, we now turn to the empirical review of studies determining whether Keynes’ Finance Circuit transmission mechanism follows proposed causal patterns.

Iheonu et al (Reference Iheonu, Asongu, Odo and Ojiem2020) investigated the first causality in the Keynes’ finance transmission mechanism, that is, the causality running from domestic bank loans to domestic investment, in ECOWAS (Economic Community of West African States) for the 1985–2017 period. Employing heterogeneous panel data methods, Iheonu et al found that domestic bank loans to the private sector Granger cause domestic investment in ECOWAS. The study recommended that domestic bank loans should be given priority when forecasting domestic investment. The present result supports the first causality relationship suggested in the Finance Circuit.

Bakari (Reference Bakari2020) examined the second sequence of the Finance Circuit on the link related to domestic investment and economic growth in Tunisia using the period between 1965 and 2016. Based on the Granger causality test, Bakari found the existence of a unidirectional causality relationship ranging from domestic investment to economic growth. In relation to policy implication, the author advised the Tunisian government to boost investment to stimulate economic growth. This result is the vindication of the second causality proposed in the Keynes Finance Circuit.

Oyedokun and Ajose (Reference Oyedokun and Ajose2018) also tested the second causality in our model related to the relationship between domestic investment and economic growth, in Nigeria for the period of 1980–2016. The Granger causality test employed, indicated that domestic investment causes economic growth in Nigeria. The study recommended that government should create an economic environment that would stimulate domestic investment through the adoption of relevant macroeconomic policies in Nigeria. The present second test also lends support to the investment economic growth causality suggested in the Financial Circuit.

Kuhe and Torruam (Reference Kuhe and Torruam2020) studied the second and the fourth links of Keynes’ finance sequential procedure using annual time series data from 1970 to 2015. Based on the Granger causality test, the study finds domestic investment to have positive and significant impact on economic growth in Nigeria. The results of the Granger causality test show statistical evidence of a bidirectional causal relationship between domestic investment and economic growth, as well as a bidirectional causal relationship between domestic savings and domestic investment. The study recommends that promoting investment for higher economic growth is an effective policy strategy for Nigeria. Investment growth through savings is also a suitable policy option in the short run, as this study shows. We conclude at this juncture that the second and the fourth links of the Keynes’ finance sequential procedure are vindicated in the present study.

Chakraborty (Reference Chakraborty2023) tested the third causal association in the Finance Circuit for BRICS countries with the help of data for the period 1990–2020. Both Granger and Dumitrescu–Hurlin panel Granger causality tests were used to explore the above direction of causality which appeared to be in favour of bidirectional causality between savings and economic growth for BRICS countries. In terms of policy implication, the author proposed that policymakers should design monetary and fiscal policies that will be either saving-friendly or income-friendly to investment, which in turn propels economic growth. The results of this study vindicate the third causal association in the Finance Circuit.

Soko (Reference Soko2023) also explored the third relationship of the Finance Circuit between South Africa’s aggregate national savings and aggregate national income from 1987 to 2021. The study confirmed that aggregate national saving was positively related to South Africa’s economic growth and found that aggregate national saving Granger caused short- and long-run economic growth. In terms of policy recommendation, the author proposed that the South African government should remove obstacles related to all efforts to mobilise national savings by implementing budgetary and monetary policies favourable to savings. The author when on to propose that high saving rates will stimulate income growth through investments in productive sectors, reducing poverty and inequality. The present study does not support the third sequence of Keynes’ monetary theory stipulating that economic growth causes savings.

Đidelija (Reference Đidelija2021) determined the direction of savings causality and economic growth in Bosnia and Herzegovina, as suggested in the third relationship of the Finance Circuit. Granger causality test and the Toda-Yamamoto procedure were applied to test the third causality link in the Finance Circuit using quarterly data from 2000 to 2016. The results of Granger causality test indicated that there is no causal link between savings and economic growth. The present result does not lend support to the Finance Circuit advocating a causal association running from economic growth to saving.

Olayiwola et al (Reference Olayiwola, Okunade and Fatai2021) also tested the third link of the Finance Circuit related to economic growth and saving based on the 2000–2019 annual data for Nigeria. Employing a multivariate Vector Error Correction Model (VECM) Granger causality test, the authors found a bidirectional causal relationship between savings and economic growth. The authors therefore urged Nigerian policymakers and the government to increase deposit rates to encourage more savings, thereby mobilising funds from the surplus side of the economy to the deficit side for productive investment. The present result does not support the link suggesting that economic growth boosts savings due to the fact that authors considered savings as a constraint to investment and not the other way round.

Hussen (Reference Hussen2020) tested the fourth causality in our model running from saving to investment. Based on data from Ethiopia for the 2000–2029 period, the author employed the Granger causality test to confirm a unidirectional causality running from investment to saving which in turn recommended an investment promoting policies to achieve better national economic performance. This study does not support the Finance Circuit causal relationship running from saving to investment.

Otoo et al (Reference Otoo, Sampson, Buabeng and Apodei2020) tested the fourth link of the Keynesian finance procedure by identifying the causal relationship between saving and investment in Ghana. The present investigation used annual time series of savings and investment in Ghana spanning from 1980 to 2017. The Johansen’s Trace and Maximum Eigenvalue tests for cointegration were performed to ascertain the level of cointegration which suggested a long-run relationship between savings and investment in Ghana. The Granger Causality test does support the fourth link in the Finance Circuit since it suggested a unidirectional causality running from savings to investment and not the other way round. In terms of policy implications, the authors recommended intensifying savings, both at the national and household level as a crucial direction for consideration if Ghana intends to finance her investments rather than relying mostly on foreign aid. However, it is crystal clear in this paper that savings cannot limit credit expansion required to finance investment.

Bukamo (Reference Bukamo2019) also investigated the last causal relationship suggested in the Keynesian finance transmission mechanism. The latter last link is related to the interaction between savings and investment. Employing Ethiopia annual time series data covering the period from 1980 to 2016, the author used Johansen cointegration test analysis to suggest a long-run relationship between savings and investment. Results found from the Granger causality test suggests bidirectional causality running from savings to investment in Ethiopia. Based on the present results, the author recommended a pursuance of policy measures towards mobilising domestic savings and to boost investment. The present investigation does not support the above fourth Finance Circuit link since the study hypothesised savings as constraint to investment and not the other way round.

The above empirical review reveals that since the publication of the General Theory in 1936 where Keynes developed an original Finance Circuit enhanced by post-Keynesians, no author has specifically undertaken a complete empirical test of the above Financial Circuit. More importantly, although causality tests related to a few Keynes’ Finance Circuit transmission mechanism patterns are investigated in the above studies, none of the above authors has specifically indicated the link of their study to Keynes or to post-Keynesian theories. The present paper, therefore, breaks new ground and aims at filling the above lacunae by undertaking in the next section a complete empirical test of the finance transmission mechanism advocated by Keynesian and post-Keynesian Economists.

Data

In the present econometric analysis, we employ annual data in our sample of 32 African countries, for the period from 1990 to 2021. These data come mainly from the World Development Indicators database (WDI, 2022) as shown in Table 1.

Table 1. Variable definitions

Source: Author.

Descriptive statistics

Table 2 presents descriptive statistics of data used in this paper. The mean and the standard deviation of variables indicate two main inferences. Firstly, the economic growth rate (GDP) is the second most stable variable in our model. This means that the GDP would be relatively clustered around the average of 1.167. Secondly, bank loans seem to be the most unstable variable in our model, which may be due to structural changes faced by African countries which consequently render the demand for money unsteady and make monetary policy implementation difficult.

Table 2. Descriptive statistics

Source: Author.

The short-run causality test

Following the seminal model of Granger (Reference Granger1969) to test the causality between two supposedly stationary variables employing time series, Hurlin (Reference Hurlin2005) and Dumitrescu and Hurlin (Reference Dumitrescu and Hurlin2012) propose to perform the latter test on a heterogeneous panel. The equation below of the autoregressive model helps to understand the meaning of their method:

$${y_{i,t}} = {\beta _i} + \mathop \sum \nolimits_{k = 1}^k {\theta _{ik}}\ {y_{i,t - k}} + \mathop \sum \nolimits_{k = 1}^k {\alpha _{ik}}\ {x_{i,t - k}} + {\varepsilon _{i,t}},\,\ \ i = 1, \ldots, N\;\;\;\;and\;\;\;t = 1, \ldots, \;T$$

Let ${y_{i,t}}$ and ${x_{i,t}}$ be stationary variables observed in N countries during T periods. The evolution of the observations between two countries implies that the coefficients on the variables relating to these countries vary. However, the lag k is assumed to be identical for all the countries in the panel. It is important to note that our model does not have random coefficients, as used by Swamy (Reference Swamy1970). The test hypotheses are such that:

$${H_0}:\;{\beta _{i1}} = \cdots = {\beta _{ik}} = 0\,\,\,\,\,\,\,\forall \;i = 1, \ldots, N$$

$${H_1}:\;{\beta _{i1}} = \cdots = {\beta _{ik}} = 0\,\,\,\,\,\,\,\,\forall \;i = 1, \ldots, {N_1}$$

$${\beta _{i1}} \ne 0\,{\rm or} \cdots {\rm or}\,{\beta _{ik}} \ne 0\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\forall {\rm{\;}}i = {N_1} + 1, \ldots, N$$

The long-run causality test

To examine the long-run causality test between our variables, we apply a panel VECM. According to Asteriou and Hall (Reference Asteriou and Hall2011); Wooldridge (Reference Wooldridge2008) and Hill et al (Reference Hill, Griffiths and Lim2011), the VECM is recommended for interconnected economic variables that maintain a long-run relationship. Thus, certain basic conditions must first be fulfilled, implying the stationarity of the same order and the cointegration of the data to be estimated; which leads us to the tests of stationarities and cointegration, before the specification of our model.

Unit root test of panel data

Levin et al (Reference Levin, Lin and Chu2002) and Im et al (2003) introduced the first-generation tests in a panel assumed to have independent errors over the periods. Pesaran et al (Reference Pesaran, Smith and Yamagata2013) for their part propose correlated errors. These tests are said to be second generation. In the case of financial data, restrictions on errors are natural in the fluctuations because of the fluidity of the data. This justifies the use of second-generation tests in the context of this study, whose equation is in the following form:

(1) $$\Delta {Y_{i,t}} = {\alpha _i} + {\varphi _i}{Y_{i,t - 1}} + {\rho _i}{\overline Y_{t - j}} + \mathop \sum \nolimits_{j = 0}^p {\theta _{ij}}{\overline Y_{t - j}} + \mathop \sum \nolimits_{j = 0}^p {\delta _{ij}}\Delta {\overline Y_t} + {\varepsilon _{i,j}}$$

where Δ is a difference operator, ${\overline Y_t} = {1 \over N}\mathop \sum \nolimits_{i = 1}^N {Y_{i,t}}$ , $\Delta {\overline Y_t} = {1 \over N}\mathop \sum \nolimits_{i = 1}^N \Delta {Y_{i,t}}$ , p denotes the number of lags, and ${\varepsilon _{i,j}}$ represents the error term. According to Palm et al (Reference Palm, Smeekes and Urbain2011), the cross-sectional means: ${\overline Y_t}$ and $\Delta {\overline Y_t}$ capture unobserved variables in Equation (1).

Cointegration tests of panel data

According to Westerlund and Basher (Reference Westerlund and Basher2008), and Westerlund et al (Reference Westerlund, Thuraisamy and Sharma2014), the tests most used in econometric analyses are those of Pedroni (Reference Pedroni1999, Reference Pedroni2004) and Kao (Reference Kao1999). However, Westerlund (Reference Westerlund2007) also proposes an alternative test relating to the structure of the error, and not to its dynamics as assumed by Pedroni and Kao.

• Pedroni Panel Cointegration Test

Pedroni (Reference Pedroni1999, Reference Pedroni2004) proposes a set of statistics (see Table 3) for testing the null hypothesis of the absence of cointegration between variables. According to Neal (Reference Neal2014), the Pedroni test allows for heterogeneity in the panel and is performed on the following equations:

(2) $${y_{it}} = {\alpha _i} + \;{\gamma _i}t + \;{\beta _{1i}}{X_{it}} + \cdots + \;{\beta _{mi}}{X_{mit}} + \;{\varepsilon _{it}}$$

(3) $$\Delta {y_{it}} = \;{\beta _{1i}}\Delta {X_{it}} + \cdots + \;{\beta _{mi}}\Delta {X_{mit}} + \;{\vartheta _{it}}$$

(4) $${\hat \varepsilon _{i,t}} = {\hat Y_i}{\hat \varepsilon _{i,t}} + \;{\hat \vartheta _{it}}$$

(5) $${\hat \varepsilon _{i,t}} = {\hat Y_i}{\hat \varepsilon _{i,t}} + {\hat \vartheta _{it}} + \mathop \sum \nolimits_{h = 1}^H {\hat Y_{i,h}}\Delta {\hat \varepsilon _{i,t - h}}{\hat \vartheta _{it}}^*$$

$i = 1, \ldots, N;\;\;t = 1, \ldots, \;T;\;m = 1, \ldots, \;M;\;h = 1, \ldots, \;H;\;{\rm{and}}\;\Delta $ represent respectively, the number of countries in the panels, the study period, the number of regressors, the number of lags, the linear trend parameter, and the difference operator. The final statistic proposed by the author is obtained at the end of the regression of each of Equations (2–5), making it possible to obtain the long-run relationship (Pedroni Reference Pedroni1999).

• The KAO Panel Cointegration Test

Table 3. Summary of stationarity and cointegration tests

Source: Author.

Note: (1) H0: No cointegration between the series of two variables. Automatic lag selection is based on AIC. (2) Trend assumption: No determistic trend. (3) Trend assumption: Deterministic intercept and trend. (4) H0: No cointegration between the series LnDPS and LnPS. Automatic lag selection is based on AIC.

*Significant at the 1 % level.

Kao (Reference Kao1999) bases his analyses on the Dickey–Fuller and augmented Dickey–Fuller tests proposed in 1981. The null hypothesis of absence of cointegration proposed by Kao (Reference Kao1999) is tested on the following model:

(6) $${y_{it}} = {\alpha _i} + {\beta _{1i}}{X_{it}} + {\varepsilon _{it}}$$

Such that $i = 1, \ldots, N;\;\;t = 1, \ldots, \;T$ designate respectively the number of countries in the panel and the observation periods. The author proposes to apply the Dickey–Fuller test to the equation:

(7) $${\hat \varepsilon _{i,t}} = \partial {\hat \varepsilon _{i,t - 1}} + \;{\mu _{i,t}}$$

${\hat \varepsilon _{i,t}}$ is estimated from Equation (7).

• Westerlund Error-correction-based Panel Cointegration Tests

On the equation:

(8) $$\Delta {y_{it}} = {d_t}{\alpha _i} + \;{\gamma _i}{Y_{i,t - 1}} + \;{\sigma _i}{X_{i,t - 1}} + \mathop \sum \nolimits_{j = 1}^J {\gamma _{i,j}}\Delta {Y_{i,t - 1}} + \mathop \sum \nolimits_{j = 1}^J {\sigma _{i,j}}\Delta {X_{i,t - 1}}\; + \;{\varepsilon _{it}}$$

Werlund suggests applying four tests (Table 3). If ${\gamma _i} \lt 0$ , there is error correction implying that the variables Y and X are cointegrated. However, when ${\gamma _i} = 0$ , this means an absence of error correction reflecting a non-cointegration between the variables tested (Westerlund Reference Westerlund2007).

• An Application of VECM Approach

The observation of long-run causality between economic variables is conditioned by the answer to the question: How to model long-run behaviour between variables? Following the work of Pédroni, Kao, and Westerlund, we propose a VECM Approach inspired by Granger (1986) in the following form:

(9) $$\Delta {y_{it}} = {\beta _{it}} + \;\mathop \sum \nolimits_{\scriptstyle i = 1 \atop \scriptstyle j = 1 } ^J {\alpha _{ij}}\Delta {Y_{i,t - k}} + {\alpha _t}\Delta CE{T_{i,t}} + \;{\mu _{it}}$$

This equation takes into account the scaled residual of the equation: ${\hat \varepsilon _{i,t}} = {Y_{it}} - {\hat \beta _i} - {\hat \alpha _i}{X_{i,t}}$ , which now contains information on the long run and the fitted process of its long-run equilibrium (Asteriou and Hall Reference Asteriou and Hall2011). CET is the corrected error term and ${\mu _{it}}$ represents the residual. Equation (9) can be specified for each frame of the economy. So, we have

• Keynes Finance Circuit Model

  (10) $$\Delta {I_{i,t}} = {\beta _{1t}} + {\alpha _{1t}}\Delta BL + {\alpha _{2t}}\Delta E{G_{i,t}} + {\alpha _{3t}}\Delta {S_{i,t}} + {\alpha _{6t}}\Delta CE{T_{i,t}}$$
  (11) $$\Delta B{L_{i,t}} = {\beta _{1t}} + {\alpha _{1t}}\Delta {I_{i,t}} + {\alpha _{2t}}\Delta E{G_{i,t}} + {\alpha _{3t}}\Delta {S_{i,t}} + {\alpha _{6t}}\Delta CE{T_{i,t}}$$
  (12) $$\Delta E{G_{i,t}} = {\beta _{1t}} + {\alpha _{1t}}\Delta B{L_{i,t}} + {\alpha _{2t}}\Delta {I_{i,t}} + {\alpha _{3t}}\Delta {S_{i,t}} + {\alpha _{6t}}\Delta CE{T_{i,t}}$$
  (13) $$\Delta {S_{i,t}} = {\beta _{1t}} + {\alpha _{1t}}\Delta B{L_{i,t}} + {\alpha _{2t}}\Delta E{G_{i,t}} + {\alpha _{3t}}\Delta {I_{i,t}} + {\alpha _{6t}}\Delta CE{T_{i,t}}$$

where EG stands for economic growth, BL stands for bank loans, S for savings, and I for investment. Two types of methods can be applied for Equations (10–13). These are the Fixed Effect Ordinary Least Squares (OLS) method, and the Generalized Method of Moments (GMM) at two levels (Eagle and Granger Reference Eagle and Granger1987).