**1. Introduction**

The power factor is a relevant measure of the efficiency of electrical energy use. A higher power factor implies efficient energy use and simultaneously ensures the safe, smooth, and efficient operation of electrical utilities. The power factor is the ratio between active power and its vector sum with reactive power, where active power executes actual work such as producing heat, illumination, and moving vehicles and machines. Reactive power is used to maintain the electromagnetic field that ensures the accomplishment of the work of active power. Summing the active and reactive power vectorially produces an amount of apparent power, which defines the total required energy capacity. A high power factor therefore implies an efficient use of the electricity generation capacity, while a low power factor reflects an abysmal use of energy. Facilities with a low power factor use more electricity than actually needed to conduct useful work [1]. A poor power factor should be corrected to reduce waste and the production costs of energy and thereby to help save the environment. There are ways to improve a low power factor to a desirable level of efficiency. The most mature, economical, and simple method is through an investment by energy consumers in capacitor banks [2]. These banks compensate for the inductive load demand of reactive power and thereby minimize the stress or burden on the electricity supply system [3].

Another alternative would be the use of synchronous alternators (or synchronous compensators). By exciting or de-exciting the magnetic fields, these inject reactive power into the network so that the voltage profiles of the system can be improved and the losses can be reduced. However, their

high installation and maintenance costs make them unsuitable for such applications in developing nations. Firms can further improve their power factor by carefully considering the design of processes and load cycles that have a repetitive nature during the design phase. Such a technique typically uses a computer algorithm to determine the ideal compromise in the relevant design parameters for improved energy efficiency and power factors. This approach, however, achieves a higher power factor at the sacrifice of reduced output levels in terms of the firms' production.

A higher power factor results in lower energy-related costs to users. According to the literature, improving power factors in industrial set-ups results in 10% to 30% cost savings. A study [4] estimated the cost savings of an independent power producer that carried out a power factor correction measure of its facility. However, the necessary investment in capacitor banks places a financial burden on users, making them reluctant to improve their power factor. Regardless, given the growing concern about climate change and increasing demand for electricity consumption, power factor correction has the potential to affect long-term economic and environmental gains to society [5]. Many countries, including Ghana, therefore urge electricity users, especially larger ones, to improve their power factor through this policy. Since 1995, the government in Ghana has imposed a policy that penalizes large-scale electricity users, labeled as special load tariff (SLT) customers, of the Electricity Company of Ghana (ECG) whose power factor falls short of 90%. This paper investigates the effectiveness of this power factor correction policy implemented by the Ghana government.

Historically, Ghana has been heavily dependent on hydro power supply, only in 2016 was it exceeded by thermal generation. Prior to that point, hydro power had accounted for the majority of the electricity supply of Ghana over history. As late as 2010, for example, the share of hydro power was approximately 70%, while that of thermal was just approximately 30%. Power from renewable energy (e.g., solar) entered the supply in 2013, but it accounts only for a negligible portion of the entire power supply [6].

In 1983, Ghana was hit by a severe drought that continued into 1984. This situation led the Akosombo Dam to experience a shortfall of water inflows, reaching below 15% of the long-term projected total [6]. This shortfall impacted the electricity supply, as the country's only electricity source was the Akosombo and Kpong hydroelectric power stations. Volta Lake did not fully recover from the 1983 drought before it was hit by another drought in 1993–1994, which again led to a reduction in electricity to consumers. This crisis exposed the country's vulnerability and security issues in terms of hydroelectric power [7].

Previous research has revealed that power factor correction measures by industrial plants, mining establishments, firms, and large commercial buildings can release 20 megawatts of tied-up electricity in a year [7]. At that time, Ghana's electricity capacity was approximately 900 megawatts, with a sizable share exported to the Republics of Togo and Benin. In addition, SLT customers represented only a part of the entire set of electricity users in Ghana. This release of 20 megawatts capacity from SLT customers was therefore considered to be relevant by the government of Ghana in that context. To improve the situation by raising the power factor among its users, in 1994, the government of Ghana announced that it would restructure tariffs to penalize those that do not adopt power factor correction measures, and it enacted this policy in January 1995 [7]. According to the policy, a user whose power factor is below a certain threshold is charged a penalty. The penalty policy is applied to large-scale electricity users, namely, the ECG's SLT customers. Tariff categories in Ghana are classified into three main groups, namely, (i) residential, (ii) nonresidential, and (iii) SLT customers [8]. Residential consumers are domestic users and nonresidential customers use electricity for commercial purposes with a capacity less than 100 kV [6]. SLT customers are defined as those that use energy for industrial purposes with loads greater than or equal to 100 kVA. According to the policy, SLT customers of ECG whose actual power factor falls short of 90% are to be penalized in proportion to the gap and to their size, measured in terms of the maximum electricity demand. For example, if the actual power factor falls short of the threshold by 5%, the penalty is 5% of the electricity bill, determined based on the maximum electricity demand of the users. Users whose power factor is above or equal to 90% are not subject to this penalty.

A relatively large body of literature has investigated the impact of energy efficiency enhancement policies on economic outcomes. For example, Costantini et al. [5] claim that sectoral energy efficiency gains display a negative effect on employment growth. They showed that this negative effect is stronger, in particular, in energy-intensive industries using data from 15 European countries. Lee and Min [9] found a negative correlation between green R&D and the financial performance of Japanese manufacturing firms. A study by Lee and Min [9] is, in the literature, one of a few that analyzes at the firm level to obtain the relationship between policy and firm economic performance.

On the one hand, these studies seem to indicate a negative relationship between energy efficiency improvement and business and economic performance. On the other hand, some of the more theory-based analyses derive opposite results. For example, by using the ASTRA model, or a dynamic, integrated macroeconomic, transport and environmental impact model, Ringel [10] concludes that enhanced green energy policies in Germany trigger tangible economic benefits in terms of GDP growth and new jobs, even in the short term. Hartwig et al. [11] use an input–output analysis-based model to show the positive growth effects and employment of energy efficiency policy in Germany. Henriques and Catarino [12] also use an input–output-based model, called the impact of sector technologies (ImSET) model, to conclude that green investment has the potential to increase employment and wage income. Allan et al. [13] use a computable general equilibrium (CGE) model to measure the impact of energy efficiency improvement in the UK.

In turn, there is a relatively smaller body of literature investigating the impact of policy on improvements in energy efficiency. Tanaka [14] provided an extensive review of energy efficiency policies implemented in the member countries of the International Energy Agency. Cox et al. [15] provided a review of non-energy- related policies. Among these studies, Xiong et al. [16] claim that a policy to restructure the industrial organization would have a large positive impact on provincial industrial energy efficiency in China. They used a slacks-based measure (SBM) that is a sophisticated variation of data-envelopment analysis (DEA), or a linear programing approach, where they allowed the existence of undesirable outputs to address the environmental burden due to inefficiency. They used a Tobit regression to reveal the positive association between industrial organization and energy efficiency as the second-stage regression following the efficiency measurement by SBM. Villca-Pozo & Gonzales-Bustos [17] found, at a provincial level, that tax policies to modernize the energy efficiency of housing in Spain have a nonsignificant impact on energy efficiency. At a more micro level, Anderson & Newell [18] found that manufacturing plants that receive government-sponsored energy audits have improved energy efficiency.

While these papers use a reduced form regression, such as the Tobit, to obtain a relationship between policy and energy efficiency, there are more model-based studies that construct a theoretical model to describe the mechanism between policy and energy use. Ringel [10] investigated the linkage between green energy policies in Germany and the country's primary energy consumption and greenhouse gases using the ASTRA model. Li et al. [19] used the VALDEX index that is a measure of energy efficiency based on value added, to find the impact of policies on eliminating low-efficiency production capacity and improving the energy efficiency of energy-intensive industries in China. Using a bottom-up approach, Fleiter et al. [20] investigated the impact of grants for small and medium enterprises in Germany to carry out energy audits of their facilities on their energy efficiency improvements.

Most of these studies, however, either showed a correlation between policy and energy efficiency or derived the results based on models that are dependent on their assumptions. For example, the Tobit regression conducted by Xiong et al. [16] shows a correlation and does not reveal the causation between policy and energy efficiency. In fact, relatively few studies have used appropriate empirical strategies to determine policy impact on energy efficiency. Yu & Zhang [21] investigated the impact of a "smart city policy" implemented in China on energy efficiency using a difference-in-differences (DID) approach at a city level. However, DID only addresses the time-invariant heterogeneity and is still based on the selections-on-observables assumption. In addition, many of these studies investigated the impacts of

energy efficiency policies at an aggregated level. To our knowledge, no study has investigated the impact of a power factor correction policy on its improvement by applying a reliable identification strategy to more micro-level data.

This paper, therefore, aims to add to the literature by identifying the causal impact of the power factor correction policy on energy efficiency improvement using firm-level data. Specifically, this study applies a regression discontinuity design (RDD), with a cutoff at the 90% penalty threshold stipulated by the policy, to five years of panel data consisting of 183 SLT customers of ECG for the years following the policy announcement in 1994 to 1997 as well as in 2012. The paper shows the effectiveness of the power factor correction policy in Ghana: firms strictly improved their power factor if it fell short of the required threshold. This paper is organized as follows. Section 2 discusses the data and identification strategies. Section 3 presents the estimation results. Finally, Section 4 concludes the paper.
