No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Machine Learning from Schools about Energy Efficiency
... By constructing counterfactual electricity consumption profiles with machine-learning forecasting methods, following the approach of Fabra et al. (2022), the study contributes to the growing literature that applies machine learning for energy analysis. Similar methodologies have been applied to estimate counterfactual consumption and analyze individual patterns in electricity use (Gonzalez-Briones et al., 2019;Burlig et al., 2020;Valentini et al., 2022). This study builds on that work, advancing the application of machine learning specifically to the context of an energy-sharing community. ...
... Recent studies have indicated that machine learning techniques enhance predictive precision for energy consumption forecasts by effectively capturing nonlinearities and complex interactions in the relationships between demand and available covariates (Gonzalez-Briones et al., 2019;Schneider et al., 2019). Machine learning is increasingly used in causal frameworks within energy economics (Burlig et al., 2020;Fabra et al., 2022), likely because the field typically fulfills the required assumptions for these estimations, like Assumption 3. Previous research has demonstrated the ability to accurately forecast energy demand using solely exogenous covariates, such as weather data (Kim andKim, 2021, Lee andCho, 2022). Moreover, concerns regarding indirect effects through prices can be alleviated, as electricity consumption is largely found to be inelastic (Ito, 2014;Fabra et al., 2021). ...
... Estimation. In a similar fashion to Burlig et al. (2020), the prediction error serves as the dependent variable. The objective is to compare machine learning predictions of electricity consumption with actual electricity use. ...
... Compared to a long event window, a short event window reduces the possibility of omitted variable bias. Methodologically, this study fits the literature that uses machine learning to support causal identification (e.g., Burlig et al. 2020;Deryugina et al. 2019;Handel and Kolstad 2017). ...
... As mentioned previously, another potential concern with the prediction approach is that the use of post-treatment data contaminates the prediction. We validate our primary findings by implementing the residual approach (Burlig et al. 2020), in which we continue to use the random forest algorithm to train the model, but with the pre-treatment data only. We Table 6 Effects of subway openings on AQI on subsample The sample includes observations during the 2 weeks surrounding each opening. ...
... To further affirm the robustness, following the approach suggested by Burlig et al. (2020), we attempt to capture uncertainty from both steps by employing a bootstrap method. Unfortunately, we find that performing the bootstrap is not feasible. ...
Public investments in subway systems are often motivated by improving local air quality. Recent studies, however, have reached different conclusions on the air quality benefits of subway investment. To reconcile these findings, this paper examines the air quality effects of all 359 subway line openings in China between 2013 and 2018. The machine learning method adopted in this paper substantially improves the consistency and precision of the estimates by purging seasonality, volatility, and the nonlinear effects of meteorological conditions in air quality data. The empirical results suggest an insignificant short-term effect and a significant long-term effect, which is expected as the adjustment of commuting mode takes time. Using the causal forest approach, the heterogeneity analysis find that a city that is experiencing rapid economic growth from a lower income level and currently has fewer subway lines is more likely to experience statistically significant improvements in air quality from a subway opening. These findings help reconcile the different findings in the literature and shed light on air pollution reduction as one of the objectives of public transit investment.
... Xia et al. (2022) used an unsupervised ML model, Fuzzy C-means algorithm to evaluate sports facilities condition in primary school. Burlig et al. (2020) evaluated school energy efficiency using ML models. Muhamedyev et al. (2020) developed a multi-criteria decision support system based on AI techniques to improve the quality of school education. ...
... Twenty-six articles utilized supervised learning approaches. As mentioned in section 1.1, supervised learning is primarily used for solving two kinds of problems, regression (predicting numerical values) and classification (predicting categorical labels) (Han et al., 2022;Janiesch et al., 2021;Sarker, 2021 (Tuba & Pelin, 2022), enrollment numbers (Zhuang & Gan, 2017), risk value of future status in online learning environment , quality score of education system (Muhamedyev et al., 2020), and specification choice in electric energy savings (Burlig et al., 2020). ...
... Four articles use only one algorithm aiming to predict the specific value and/or identify the key factors that influence that value. For example, Burlig et al. (2020) found the electricity specification choice set with a central estimate of 60% would contribute a great energy savings by using a Linear Regression algorithm; Through Gradient Boosting algorithm, Muhamedyev et al. (2020) found location of the school, higher category teachers, more readers in the library, sports facilities, technical support and other clubs would contribute to schools of good quality. Masci et al. (2018) found that in almost all the countries, the most important variables that influenced math scores included students' self-reported anxiety toward tests, socioeconomic, and self-reported motivation. ...
Background: The use of Artificial Intelligence (AI) has increased in all education sectors including K-12 settings
where students can learn about AI and have an augmented learning experience using AI.
Purpose: The purpose of this systematic review is to provide a more complete and nuanced understanding of the
role and impact of AI in K-12 education by synthesizing publication trends, AI research themes, AI methods and
technology applications, and AI use by students and teachers in K-12 educational settings.
Methods: The systematic review searched Web of Science and six databases indexed in EBSCO host. A PRISMA
flow chart was applied to search and screen for studies. Articles were screened at the title, abstract and full-text
level and coded and analyzed.
Results: Themes in 66 AI studies include AI as a predictor and indicator of academic behavior or performance, AI
curriculum design, integrating AI in various subjects, evaluation of AI in education, AI to enhance learning
environments and school operations, AI ethics, and the equity and safety of AI. AI methods were grouped into
Supervised Learning, Unsupervised Learning and Reinforcement Learning. AI technology applications were
Machine Learning (ML) model building tools, intelligent tutors, chat bot, educational games, AI robots and
virtual reality devices. AI applications were mostly used by teachers for ML model demonstration, academic
performance prediction and behavior prediction. AI was used by students for scientific discovery learning,
improving learning experience and data driven decisions.
Conclusion: This review has implications for K-12 school personnel and researchers. Practitioners can use the
findings to implement AI in K-12 education. Researchers can benefit from the findings of the review but also
build on the gap in research on AI K-12 education.
... Varian (2016) suggested that counterfactual building is essentially a predictive task for which ML is very well suited. Early literature with empirical applications using ML in a counterfactual setting has appeared (Abrell et al., 2019;Benatia & Gingras, 2022;Burlig et al., 2020;Cerqua & Letta, 2022;Cerqua et al., 2021;Resce, 2022;Souza, 2019). Abrell et al. (2019), Burlig et al. (2020) and Souza (2019) are in the energy economics field. ...
... Early literature with empirical applications using ML in a counterfactual setting has appeared (Abrell et al., 2019;Benatia & Gingras, 2022;Burlig et al., 2020;Cerqua & Letta, 2022;Cerqua et al., 2021;Resce, 2022;Souza, 2019). Abrell et al. (2019), Burlig et al. (2020) and Souza (2019) are in the energy economics field. Abrell et al. (2019), like our approach, can eliminate the bias in the absence of a control group. ...
... In their study, however, they focus on the impact of a continuous treatment rather than a discrete treatment like the index suspension. Burlig et al. (2020) make use of an untreated control group to eliminate the prediction bias caused by the prediction algorithm. Souza (2019) can rely on an original control group. ...
This paper analyzes the impact of automatic wage indexation on employment. To boost competitiveness and increase employment, Belgium suspended its automatic wage indexation system in 2015. This resulted in a 2% fall in real wages for all workers. In the absence of a suitable control group, we use machine learning for the counterfactual analysis. We artificially construct the control group for a difference-in-difference analysis based on the pre-treatment evolution of treated firms. We find a positive impact on employment of 1.2%, which corresponds to a labor demand elasticity of − 0.6. This effect is more pronounced for manufacturing firms, where the elasticity reaches − 1. These results show that a suspension of the automatic wage indexation mechanism can be effective in preserving employment.
... Recently, there have been initial practical applications of this counterfactual methodology in empirical studies (Benatia, 2020;Benatia and de Villemeur, 2019;Bijnens et al., 2019;Burlig et al., 2020;Cerqua et al., 2021;Souza, 2019). The majority of these research efforts encounter a common challenge, which is the absence of an original control group, similar to our case. ...
... The approach employed in this article to evaluate the impact of the COVID-19 pandemic involves constructing a counterfactual scenario using a Machine Learning model, as indicated by (Cerqua et al., 2021;Varian, 2016;Burlig et al., 2020;Cerqua and Letta, 2022). The term "Machine Learning Control Method" (MLCM), introduced by Cerqua et al. (2021), denotes this particular approach. ...
This study delves into the impact of the COVID-19 pandemic on the enrollment rates of on-site undergraduate programs within Brazilian public universities. Employing the Machine Learning Control Method, a counterfactual scenario was constructed in which the pandemic did not occur. By contrasting this hypothetical scenario with real-world data on new entrants, a variable was defined to characterize the impact of the COVID-19 pandemic on on-site undergraduate programs at Brazilian public universities. This variable reveals that the impact factor varies significantly when considering the geographical locations of the institutions offering these courses. Courses offered by institutions located in smaller population cities experienced a more pronounced impact compared to those situated in larger urban centers.
... The empirical framework based on machine learning counterfactual predictions used in this paper was inspired by the work of Burlig et al. (2020) on energy efficiency. There is a burgeoning literature in energy and environmental economics using machine learning methods for policy evaluation and regulation (Abrell et al., 2019;Benatia and Billette de Villemeur, 2019;Benatia, 2022;Fabra et al., 2022;Graf et al., 2020). ...
... They have a causal interpretation as the effects of containment measures on electricity consumption. This method, inspired by Burlig et al. (2020), allows performing causal inference at the hourly level. Its main advantage compared to standard regression methods is that it yields more accurate hourly predictions to be used as inputs in a structural model of hourly prices, as is done in Section 3. 16 Obviously, many other machine learning methods could be used for this same purpose. ...
The COVID-19 crisis has disrupted electricity systems worldwide. This article disentangles the effects of the demand reductions, fuel price devaluation, and increased forecast errors on New York’s day-ahead and real-time markets by combining machine learning and structural econometrics. From March 2020 to February 2021, statewide demand has decreased by 4.6 TWh (-3%) including 4 TWh (-8%) for New York City alone, and the day-ahead market has depreciated by 15 million (+23%) because of abnormally large forecast errors which significantly undermined system efficiency.
... Estimates of the average impact on firms that received treatment show that our intervention reduced firms' average unit cost of electricity but did not have a statistically significant effect on electricity use. We focus on electricity use as the main outcome variable instead of an intensity measure following recent experimental and quasi-experimental studies of energy efficiency 19,[37][38][39] . These studies note that accurate output/service level measures for calculating energy intensity are usually unavailable, and poorly measured proxies can confound the estimation of the relationships of interest, which we expect to be similarly true in our setting. ...
Increasing the efficiency of industrial energy use is widely considered important for mitigating climate change. We randomize assignment of an energy audit intervention aimed at improving energy efficiency and reducing energy expenditures of small- and medium-sized metal processing firms in Shandong Province, China, and examine impacts on energy outcomes and interactions with firms’ management practices. We find that the intervention reduced firms’ unit cost of electricity by 8% on average. Firms with more developed structured management practices showed higher rates of recommendation adoption. However, the post-intervention electricity unit cost reduction is larger in firms with less developed practices, primarily driven by a single recommendation that corrected managers’ inaccurate reporting of transformer usage at baseline, lowering their electricity costs. By closing management-associated gaps in awareness of energy expenditures, energy audit programmes may reduce a firm’s unit cost of energy but have an ambiguous impact on energy use and climate change.
... The ecoEnergy Retrofit Homes program (ecoEnergy) was announced by the Canadian federal government Our use of assessed value as a proxy for wealth in Canada is explained further in the following Sections. 6 Other studies have documented that building simulation model predictions of retrofit energy savings fall short of observed reductions across common energy efficiency investments, such as (Fowlie et al., 2018;Burlig et al., 2020;Christensen et al., 2021;Chuang et al., 2022;Alekhanova et al., 2023). 7 During part of this period, houses were also eligible to apply for the Home Renovation Tax Credit, and some provinces offered home retrofit incentives that piggy-backed on the federal program (Rivers and Shiell, 2016 launched the Greener Homes program, which has a budget of $2.6 billion. ...
Maintaining household welfare in the transition to a net zero economy is critical to the public acceptance of climate policy. A challenge in meeting this goal is our incomplete understanding of the distribution of household-level benefits from policies designed to reduce greenhouse gases in residential buildings. We provide new insights on key variables that contribute to household and social welfare by quantifying both the level and distribution of energy savings, bill savings, and rebates disbursed from Canada’s national energy efficiency retrofit program. Using a unique dataset consisting of electricity and natural gas consumption from all single-family homes in a Canadian city, we find that adopted retrofits reduce natural gas consumption for up to 10 years in the average participating house by about 20% and whole-envelope retrofits reduce natural gas consumption by 35%. However, these savings
represent only about half of pre-retrofit predicted savings, and several recommended retrofits save zero energy. While energy bill savings exhibit a modest peak among some lower wealth properties, retrofit rebates were disbursed equally across the house wealth distribution.
... The first wave of empirical research found that rebound effects due to improvements in energy efficiency offset a small fraction of potential savings but rarely caused Jevons's paradox (Sorrell et al. 2009;Gillingham et al. 2013;Chan and Gillingham 2015). A more recent strand of empirical work emphasizes the role of technological heterogeneities and biased efficiency estimates (Fowlie et al. 2018;Burlig et al. 2020;Christensen et al. 2023). Alpizar et al. (2024) use a randomized control trial to estimate the dynamic treatment effects of households receiving water-efficient technologies. ...
As global aquifer levels continue to decline, clarifying the impact of irrigation efficiency improvements on water resources is critically important. This study uses two transitions in irrigation technology to investigate whether rebound effects cause such efficiency improvements to increase resource extraction, a phenomenon known as Jevons’s paradox. We demonstrate how staggered adoption of an irrigation technology and dynamic treatment effects cause two-way fixed effects (TWFE) to indicate the wrong sign for the effect on withdrawals. Using an estimator appropriate for these circumstances, we find no significant evidence of Jevons’s paradox. The dynamic effects we find explain this discrepancy and, perhaps more important, reveal irrigators’ process of adaptation to each new technology at the intensive and extensive margins.
... A potential solution in energy-constrained networks is to focus on quality of service (QoS) when minimizing energy consumption is possible. To meet all requests requiring energy-efficient computation, records accumulated throughout the system are the foundation for higher-layer resolution [20] and [21]. Using a long-distance remote cloud server for delay-sensitive applications increases interruptions and decreases QoS. ...
Connecting to the internet will increase our computing challenges as it becomes an integral part of our daily lives. Therefore, it is necessary to advance the service qualities of Internet of Things (IoT) applications because the data produced by all these devices will need to be processed quickly and sustainably. Previously, cloud data centers with large capacity interface IoT devices with support servers. While IoT devices proliferate and generate massive amounts of data, communicating between devices and the Cloud is becoming more complex and harder, resulting in high costs and inefficiencies. Fog computing emerges as an approach to address the growing demand for IoT solutions. In this article, an IoT-fog-cloud application's general framework is developed, followed by an algorithm for Energy efficiency through an integrated approach computation model. Fog-Enabled Smart Cities (FESC) are proposed to minimize service delay and response time by using a fog offloading policy for the fog-enabled IoTs. Also, we developed an analytical model evaluating the proposed framework's effectiveness in reducing the delay of IoT services. Comparing the proposed model and the Alternating Direction Method of Multipliers (ADMM-VS) algorithm, the proposed model performs significantly better. Thus, by optimizing response and processing times, fog-enabled smart grids determine whether computation will be performed autonomously or semi-autonomously on fog nodes or in the Cloud.
... In one example,Ruz, Varas, and Villena (2013) use k-means clustering algorithms to identify the common characteristics of households lacking internet access as a means of evaluating whether an unconditioned broadband and subsidiary campaign had a significant effect on broadband penetration in Chile.Zheng, Zheng, and Ye (2016) also use machine learning methods to assess the development impact of environmental tax reform in China. Niu, Wang, and Duan(2009)rely on support vector machine analysis to evaluate the impact of power plant construction projects in China, andBurlig et al. (2017) examine, via machine learning, the impact of energy efficiency upgrades in primary and secondary schools. Machine learning can also yield useful meta-analytical insights.Mueller, Gaus, and Konradt (2016) note that progress in evaluation research depends on establishing a productive cycle of scholarly knowledge generation, dissemination, and implementation. ...
... (2019) uses a ML approach to measure the costs of air pollution. Burlig et al. (2020) use ML to refine estimates of energy efficiency improvements. Stetter, Mennig, and Sauer (2022) use a DML approach to measure effectiveness of an agricultural intervention, and Klosin and Vilgalys (2022) introduce a DML approach to measure elasticities in a panel setting. ...
I measure adaptation to climate change by comparing elasticities from short-run and long-run changes in damaging weather. I propose a debiased machine learning approach to flexibly measure these elasticities in panel settings. In a simulation exercise, I show that debiased machine learning has considerable benefits relative to standard machine learning or ordinary least squares, particularly in high-dimensional settings. I then measure adaptation to damaging heat exposure in United States corn and soy production. Using rich sets of temperature and precipitation variation, I find evidence that short-run impacts from damaging heat are significantly offset in the long run. I show that this is because the impacts of long-run changes in heat exposure do not follow the same functional form as short-run shocks to heat exposure.
... A particular appeal of these methods is the ability to predict counterfactuals in order to test for causality. This method, combined with existing econometric techniques, has been used to examine treatment effects for energy efficiency upgrades in schools, outperforming standard panel-fixed effects approaches [170]. Additionally, ML methods are useful for estimating heterogenous treatment effects and have been applied by a number of researchers in this regard, in particular, on high-dimensional smart metering datasets [171], [172]. ...
... 6 An earlier working paper version of this work dated from 2022 does not include customer-level energy price data, metrics on household bill savings from energy efficiency retrofits, rebates received by households, or the distributional analysis. 7 Other studies have documented that building simulation model predictions of retrofit energy savings fall short of observed reductions across common energy efficiency investments, such as (Fowlie et al., 2018;Burlig et al., 2020;Christensen et al., 2021;Chuang et al., 2022;Alekhanova et al., 2023). 8 During part of this period, houses were also eligible to apply for the Home Renovation Tax Credit, and some provinces offered home retrofit incentives that piggy-backed on the federal program (Rivers and Shiell, 2016). ...
Over 50 countries representing three quarters of global CO 2 emissions have pledged to achieve a net-zero carbon economy by 2050, and energy efficiency improvements are a primary contributor in policy scenarios that attain this goal. However, uncertainties remain about the realized effectiveness of energy efficiency programs. This paper provides evidence on the realized savings from Canada's largest residential energy retrofit program. We use utility data from all single-family homes in a mid-sized Canadian city and detailed energy audit records from the EnerGuide for Homes database, which includes modeled predictions of energy savings from retrofit adoptions. The retrofit program reduces natural gas consumption in the average participating home by about 21%, representing 60% of predicted natural gas savings. Whole-envelope retrofits are predicted to reduce natural gas consumption by 67%, but in practice only half of these savings are realized. This underscores the importance of developing new modeling approaches that incorporate house-level utility data, which reflect the outcome of realized rather than predicted occupant behavior, to increase retrofit energy savings and return per subsidy dollar spent.
... Most of them, such as causal forests (Wager & Athey, 2018), and matrix completion methods for panel data (Athey et al., 2021) require the availability of a control group in order to plausibly estimate causal effects. Other applications use ML to forecast a counterfactual scenario, but still in the traditional setting with a control group (Burlig et al., 2020;1 Another approach similar to ITS, but not originating in the time series literature, is the regression discontinuity in time, which derives from an adaptation of the regression discontinuity framework to applications where time is the running variable and treatment begins at a particular threshold in time (Anderson, 2014). This approach is, however, of limited applicability, as it requires the availability of -at least -tens of time periods before and after treatment, and has been recently criticized for the much stronger assumptions underpinning it, compared to the classical version of the regression discontinuity design (see Hausman and Rapson, 2018). 2 This is because in ITS designs the effect is usually given by an estimated coefficient, e.g., under the assumption that the intervention produced a level shift on the outcome, it is the coefficient of a binary dummy variable that is 1 in the postintervention period and 0 otherwise. ...
The standard way of estimating treatment effects relies on the availability of a similar group of untreated units. Without it, the most widespread counterfactual methodologies cannot be applied. We tackle this limitation by presenting the Machine Learning Control Method (MLCM), a new causal inference technique for aggregate data based on counterfactual forecasting via machine learning. The MLCM is suitable for the estimation of individual, average, and conditional average treatment effects in evaluation settings with short panels and no controls. The method is formalized within the Rubin's Potential Outcomes Model and comes with a full set of diagnostic, performance, and placebo tests. We illustrate our methodology with an empirical application on the short-run impacts of the COVID-19 crisis on income inequality in Italy, which reveals a striking heterogeneity in the inequality effects of the pandemic across the Italian local labor markets.
... To verify the success or failure of energy efficiency interventions, statistical ML offers methods for causal inference. For example, Burlig et al. [102] used Lasso regression on hourly electricity consumption data from schools in California to find that energy efficiency interventions fall short of the expected savings. Such problems could represent a useful application of deep learning methods for counterfactual prediction [332]. ...
Climate change is one of the greatest challenges facing humanity, and we, as machine learning (ML) experts, may wonder how we can help. Here we describe how ML can be a powerful tool in reducing greenhouse gas emissions and helping society adapt to a changing climate. From smart grids to disaster management, we identify high impact problems where existing gaps can be filled by ML, in collaboration with other fields. Our recommendations encompass exciting research questions as well as promising business opportunities. We call on the ML community to join the global effort against climate change.
... Machine learning (ML) has been widely used in the modeling and prediction of power systems because it allows the simulation of offline behavior and anticipates failures. This has led to a remarkable increase in the precision, robustness, and ability to generalize the behavior of these systems [31][32][33]. The models obtained with ML allow forecasting of the energy consumption and performance of buildings, and although there are many techniques for determining these models, these all provide reasonable accuracy by providing a large amount of data and optimizing the parameters [6,34,35]. ...
The intelligent analysis of electrical parameters has been facilitated by the Internet of Things (IoT), with capabilities to access a lot of data with customized sampling times. On the contrary, binary classifiers using support vector machines (SVM) resolve nonlinear cases through kernel functions. This work presents two binary classifiers of presence in the home using total household active power data obtained from the automated reading of an IoT device. The classifiers consisted of SVM using kernel functions, a linear function, and a nonlinear function. The data was acquired with the Emporia Gen 2 Vue energy monitor for 20 days without interruption, obtaining averaged readings every 15 min. Of these data, 75% was for training the classifiers, and the rest of the data was for validation. Contrary to expectations, the evaluation yielded accuracies of 91.67% for the nonlinear SVM and 92.71% for the linear SVM, concluding that there was similar performance.
... When policymakers face budget constraints, identifying those who should be treated is critical to maximizing policy impacts. Advances in machine learning and econometric methods have led to a surge in research on targeting in many policy domains, including job training programs (Kitagawa and Tetenov, 2018), social safety net programs (Finkelstein and Notowidigdo, 2019;Deshpande and Li, 2019), energy efficiency programs (Burlig, Knittel, Rapson, Reguant, and Wolfram, 2020), behavioral nudges for electricity conservation (Knittel and Stolper, 2021), and dynamic electricity pricing (Ito, Ida, and Takana, forthcoming). ...
... The K-Means algorithm has been implemented by García et al. [31] to perform automatic customer clustering to split the residential and non-residential customers based on the behaviour during the pandemic. Burlig et al. [32] generated the counterfactual outcome with the Lasso regression model for analysing the energy efficiency, then comparing the predicted outcome to the realised outcome to calculate the treatment effect. Graf et al. [30] proposed a counterfactual neural network model to predict businessas-usual re-dispatch costs in the electricity market and compare them to actual re-dispatch costs for the pre-covid and Covid-19 lockdown periods. ...
The electricity industry is heavily implementing smart grid technologies to improve reliability, availability, security, and efficiency. This implementation needs technological advancements, the development of standards and regulations, as well as testing and planning. Smart grid load forecasting and management are critical for reducing demand volatility and improving the market mechanism that connects generators, distributors, and retailers. During policy implementations or external interventions, it is necessary to analyse the uncertainty of their impact on the electricity demand to enable a more accurate response of the system to fluctuating demand. This paper analyses the uncertainties of external intervention impacts on electricity demand. It implements a framework that combines probabilistic and global forecasting models using a deep learning approach to estimate the causal impact distribution of an intervention. The causal effect is assessed by predicting the counterfactual distribution outcome for the affected instances and then contrasting it to the real outcomes. We consider the impact of Covid-19 lockdowns on energy usage as a case study to evaluate the non-uniform effect of this intervention on the electricity demand distribution. We could show that during the initial lockdowns in Australia and some European countries, there was often a more significant decrease in the troughs than in the peaks, while the mean remained almost unaffected.
... Many scholars have adopted machine learning algorithms to predict the economic growth [3] [4]. For example, Nang et al. [5] applied Support Vector Machines (SVM) to the regression of GDP and obtained some results. ...
This article investigates the incentives for firms with market power to manipulate markets by strategically reneging on forward commitments. We first study the behavior of a dominant firm in a two‐period model with demand uncertainty. We then use the model's predictions and a machine learning approach to investigate multiple occurrences of reneging on long‐term commitments in Alberta's electricity market in 2010–2011. We find that a supplier significantly increased its revenues by strategically reneging on its capacity availability obligations, causing Alberta's annual electricity procurement costs to increase by as much as $600 million (+17%).
Firms expect to recover the fixed costs required to start production by earning positive operating profits in subsequent periods. We develop a dynamic competitive benchmark that accounts for start-up costs, showing that static markups overstate the rents attributable to market power in an electricity market where generators frequently stop and start production in response to rooftop solar output. We demonstrate that the large-scale expansion of solar capacity can lead to increases in the collective profitability of fossil fuel plants because competition softens at sunset—plants displaced by solar during the day must incur start-up costs to compete in the evening. (JEL D24, L13, L25, L94, Q41, Q42)
The advent of herbicide-tolerant genetically modified (GM) crops spurred rapid and widespread use of the herbicide glyphosate throughout US agriculture. In the two decades following GM-seeds’ introduction, the volume of glyphosate applied in the United States increased by more than 750%. Despite this breadth and scale, science and policy remain unresolved regarding the effects of glyphosate on human health. We identify the causal effect of glyphosate exposure on perinatal health by combining 1) county-level variation in glyphosate use driven by 2) the timing of the GM technology and 3) differential geographic suitability for GM crops. Our results suggest the introduction of GM seeds and glyphosate significantly reduced average birthweight and gestational length. While we find effects throughout the birthweight distribution, low expected-weight births experienced the largest reductions: Glyphosate’s birthweight effect for births in the lowest decile is 12 times larger than that in the highest decile. Together, these estimates suggest that glyphosate exposure caused previously undocumented and unequal health costs for rural US communities over the last 20 years.
In many settings, misaligned incentives and inadequate monitoring lead employees to take self-interested actions. This paper identifies and quantifies the costs of this principal-agent problem in the context of an energy efficiency appliance replacement program. I show that contractors (agents) hired by the electric utility (the principal) increase their compensation by intentionally misreporting program data to authorize the replacement of nonqualified refrigerators. I estimate that each unqualified replacement reduces program benefits by 60 per replacement. (JEL D82, L68, L94, L98)
The electricity industry is heavily implementing intelligent control systems to improve reliability, availability, security, and efficiency. This implementation needs technological advancements, the development of standards and regulations, as well as testing and planning. Load forecasting and management are critical for reducing demand volatility and improving the market mechanism that connects generators, distributors, and retailers. During policy implementations or external interventions, it is necessary to analyse the uncertainty of their impact on the electricity demand to enable a more accurate response of the system to fluctuating demand. This paper analyses the uncertainties of external intervention impacts on electricity demand. It implements a framework that combines probabilistic and global forecasting models using a deep learning approach to estimate the causal impact distribution of an intervention. The causal effect is assessed by predicting the counterfactual distribution outcome for the affected instances and then contrasting it to the real outcomes. We consider the impact of Covid-19 lockdowns on energy usage as a case study to evaluate the non-uniform effect of this intervention on the electricity demand distribution. We could show that during the initial lockdowns in Australia and some European countries, there was often a more significant decrease in the troughs than in the peaks, while the mean remained almost unaffected.
A survey of virtual and augmented reality for enhancing learning perspective
We investigate the impact of Daylight Saving Time (DST) on electricity demand in the Balearic Islands (Spain), a region characterized by a large tourism industry, and where tourists outnumber residents during the peak season. Using daily data from 2007 to 2017 and deploying a time series model as well as a double/debiased machine learning model, we estimate that DST implementation results in a small positive increase of electricity consumption, although in several cases this effect is not statistically significant. Omitting from the model a variable that measures the number of people present on the islands would lead to an overestimation of the DST effect on electricity consumption.
We study an energy efficient charcoal cookstove in an experiment with 1,000 households in Nairobi. We estimate a 39 percent reduction in charcoal spending, which matches engineering estimates, generating a 295 percent annual return. Despite fuel savings of 295 in emissions reductions—households are only willing to pay $12. Drawing attention to energy savings does not increase demand. However, a loan more than doubles willingness to pay: credit constraints prevent adoption of privately optimal technologies. Energy efficient technologies could drive sustainable development by slowing greenhouse emissions while saving households money. (JEL D12, D91, G51, O12, O13, O32, Q54)
The teacher is not possible to teach all those things among their students and there is a possibility
for the teacher to engage their students into different self learning techniques. At present digital age
observes educators attention to the modifications on characteristics of learners in the learning environment.
There are different types of technology resources which the learners using for the improvement of their
learning. The period of globalization academic performance of students has closely connected with
different types of learning method. At present there are different learning methods are available however,
the few learning methods are very applicable for the students who improve their academic performance.
The self directed learning method is most important method for the young minds of toady. At present the
culture of learning among the learners gradually moves to self-directed learning because it is a self paced
one whenever the learners have interest who can able to learn. The self directed learning method is most
effective method especially young minds of today. It provides self independency among the learners.
Hence, the teacher of today motivates the young minds of today to involve the self directed learning.
Many countries have phased out nuclear power in response to concerns about nuclear waste and the risk of nuclear accidents. This paper examines the shutdown of more than half of the nuclear production capacity in Germany after the Fukushima accident in 2011. We use hourly data on power plant operations and a machine learning approach to estimate the impacts of the phase-out policy. We find that reductions in nuclear electricity production were offset primarily by increases in coal-fired production and net electricity imports. Our estimates of the social cost of the phase-out range from €3 to €8 billion per year. The majority of this cost comes from the increased mortality risk associated with exposure to the local air pollution emitted when burning fossil fuels. Policymakers would have to significantly overestimate the risk or cost of a nuclear accident to conclude that the benefits of the phase-out exceed its social costs. We discuss the likely role of behavioral biases in this setting, and highlight the importance of ensuring that policymakers and the public are informed about the health effects of local air pollution.
The ambitious climate targets set by industrialized nations worldwide cannot be met without decarbonizing the building stock. Using Germany as a case study, this paper takes stock of the extensive set of energy efficiency policies that are already in place and clarifies that they have been designed “in good faith” but lack in overall effectiveness as well as cost-efficiency in achieving these climate targets. We map out the market failures and behavioural considerations that are potential reasons for why realized energy savings fall below expectations and why the household adoption of energy-efficient and low-carbon technologies has remained low. We highlight the pressing need for data and modern empirical research to develop targeted and cost-effective policies seeking to correct these market failures. To this end, we identify some key research questions and identify gaps in the data required for evidence-based policy.
We estimate the effects of temperature on human capital production in India. We show that high temperatures reduce math and reading test scores among school-age children. Agricultural income is one mechanism driving this relationship: hot days during the growing season reduce agricultural yields and test scores with comparatively modest effects of hot days in the non-growing season. The roll-out of a workfare program, by providing a safety net for the poor, substantially weakens the link between temperature and test scores. Our results imply that absent social protection programs, higher temperatures will have large negative impacts on human capital production of poor populations in agrarian economies.
Energy-efficient technologies offer considerable promise for reducing the financial costs and environmental damages associated with energy use, but it has long been observed that these technologies may not be adopted by individuals and firms to the degree that might be justified, even on a purely financial basis. We survey the relevant literature on this "energy-efficiency gap" by presenting two complementary frameworks. First, we divide potential explanations for the energy-efficiency gap into three categories: market failures, behavioral explanations, and model and measurement errors. Second, we organize previous research in terms of the fundamental elements of cost-minimizing energy-efficiency decisions. This provides a decomposition that organizes thinking around four questions. First, are product offerings and pricing economically efficient? Second, are energy operating costs inefficiently priced and/or understood? Third, are product choices cost minimizing in present value terms? Fourth, do other costs inhibit more energy-efficient decisions? We synthesize academic research on these questions, with an emphasis on recent empirical findings, and offer suggestions for future research. ( JEL D24, D82, L94, L98, O33, Q41, Q48).
The covariate-balancing propensity score (CBPS) extends logistic regression to simultaneously optimize covariate balance and treatment prediction. Although the CBPS has been shown to perform well in certain settings, its performance has not been evaluated in settings specific to pharmacoepidemiology and large database research. In this study, we use both simulations and empirical data to compare the performance of the CBPS with logistic regression and boosted classification and regression trees. We simulated various degrees of model misspecification to evaluate the robustness of each propensity score (PS) estimation method. We then applied these methods to compare the effect of initiating glucagonlike peptide-1 agonists versus sulfonylureas on cardiovascular events and all-cause mortality in the US Medicare population in 2007-2009. In simulations, the CBPS was generally more robust in terms of balancing covariates and reducing bias compared with misspecified logistic PS models and boosted classification and regression trees. All PS estimation methods performed similarly in the empirical example. For settings common to pharmacoepidemiology, logistic regression with balance checks to assess model specification is a valid method for PS estimation, but it can require refitting multiple models until covariate balance is achieved. The CBPS is a promising method to improve the robustness of PS models.
Commercial buildings play a major role in determining U.S. greenhouse gas emissions, yet surprisingly little is known about the environmental performance of different buildings at a point in time or how the same buildings perform over time. By exploiting a unique panel of commercial buildings from a major electric utility, we study the association between a building’s electricity consumption and the physical attributes of buildings, lease incentive terms, indicators of human capital, and climatic conditions. We find that buildings that are newer and of higher quality consume more electricity, contrasting evidence for the residential sector. However, using our panel data set, we document that newer buildings are most resilient when exposed to hotter weather. Those buildings that have a building manager on site and whose tenants face a positive marginal cost for electricity also demonstrate a better environmental performance.
We develop projections of future spending on, and savings from, energy efficiency programs funded by electric and gas utility customers in the USA, under three scenarios through 2025. Our analysis, which updates a previous LBNL study, relies on detailed bottom-up modeling of current state energy efficiency policies, regulatory decisions, and demand-side management and utility resource plans. The three scenarios are intended to represent a range of potential outcomes under the current policy environment (i.e., without considering possible major new policy developments).
Key findings from the analysis are as follows:
By 2025, spending on electric and gas efficiency programs (excluding load management programs) is projected to double from 2010 levels to 15.6 billion in the high case and $6.5 billion in the low case.
Compliance with statewide legislative or regulatory savings or spending targets is the primary driver for the increase in electric program spending through 2025, though a significant share of the increase is also driven by utility DSM planning activity and integrated resource planning.
Our analysis suggests that electric efficiency program spending may approach a more even geographic distribution over time in terms of absolute dollars spent, with the Northeastern and Western states declining from over 70 % of total USA spending in 2010 to slightly more than 50 % in 2025, and the South and Midwest splitting the remainder roughly evenly.
Under our medium case scenario, annual incremental savings from customer-funded electric energy efficiency programs increase from 18.4 TWh in 2010 in the USA (which is about 0.5 % of electric utility retail sales) to 28.8 TWh in 2025 (0.8 % of retail sales).
These savings would offset the majority of load growth in the Energy Information Administration’s most recent reference case forecast, given specific assumptions about the extent to which future energy efficiency program savings are captured in that forecast.
The pathway that customer-funded efficiency programs ultimately take will depend on a series of key challenges and uncertainties associated both with the broader market and policy context and with the implementation and regulatory oversight of the energy efficiency programs themselves.
Research on climate change suggests that small improvements in the "sustainability" of buildings can have large effects on greenhouse gas emissions and on energy efficiency in the economy. This paper analyzes the economics of "green" building. First, we analyze a panel of office buildings "certified" by independent rating agencies, finding that large recent increases in the supply of green buildings and the unprecedented volatility in property markets have not significantly affected the relative returns to green buildings. Second, we analyze a large cross section of office buildings, demonstrating that economic premiums in rent and asset values are substantial. Third, we relate the economic premiums for green buildings to their sustainability, confirming that the attributes rated for both thermal efficiency and sustainability contribute to premiums in rents and asset values. Even among green buildings, increased energy efficiency is fully capitalized into rents and asset values.
Building on an idea in Abadie and Gardeazabal (2003), this article investigates the application of synthetic control methods to comparative case studies. We discuss the advantages of these methods and apply them to study the effects of Proposition 99, a large-scale tobacco control program that California implemented in 1988. We demonstrate that following Proposition 99 tobacco consumption fell markedly in California relative to a comparable synthetic control region. We estimate that by the year 2000 annual per-capita cigarette sales in California were about 26 packs lower than what they would have been in the absence of Proposition 99. Given that many policy interventions and events of interest in social sciences take place at an aggregate level (countries, regions, cities, etc.) and affect a small number of aggregate units, the potential applicability of synthetic control methods to comparative case studies is very large, especially in situations where traditional regression methods are not appropriate. The methods proposed in this article produce informative inference regardless of the number of available comparison units, the number of available time periods, and whether the data are individual (micro) or aggregate (macro). Software to compute the estimators proposed in this article is available at the authors' web-pages.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org.
Most analyses of energy efficiency investments ignore that the value of electricity varies widely across hours. We show how much timing matters. Using novel hourly consumption data from an air conditioner rebate program in California, we find that energy savings are concentrated in high-value hours. This significantly increases the value of these investments, especially after we account for the large capacity payments that electricity generators receive to guarantee supply in peak hours. We then use engineering predictions to calculate timing premiums for a wide range of energy efficiency investments, finding substantial variation in economic value across investments. (JEL L94, L98, Q41, Q48)
Many settings in empirical economics involve estimation of a large number of parameters. In such settings, methods that combine regularized estimation and data-driven choices of regularization parameters are useful. We provide guidance to applied researchers on the choice between regularized estimators and data-driven selection of regularization parameters. We characterize the risk and relative performance of regularized estimators as a function of the data-generating process and show that data-driven choices of regularization parameters yield estimators with risk uniformly close to the risk attained under the optimal (unfeasible) choice of regularization parameters. We illustrate using examples from empirical economics.
A growing number of policies and programs aim to increase investment in energy efficiency, because conventional wisdom suggests that people fail to take up these investments even though they have positive private returns and generate environmental benefits. Many explanations for this energy efficiency gap have been put forward, but there has been surprisingly little field testing of whether the conventional wisdom is correct. This article reports on the results of an experimental evaluation of the nation's largest residential energy efficiency program- theWeatherization Assistance Program-conducted on a sample of approximately 30,000 households in Michigan. The findings suggest that the upfront investment costs are about twice the actual energy savings. Furthermore, the model-projected savings are more than three times the actual savings. Although this might be attributed to the "rebound" effect-when demand for energy end uses increases as a result of greater efficiency-the article fails to find evidence of significantly higher indoor temperatures at weatherized homes. Even when accounting for the broader societal benefits derived from emissions reductions, the costs still substantially outweigh the benefits; the average rate of return is approximately -7.8% annually. © The Author(s) 2018. Published by Oxford University Press on behalf of the President and Fellows of Harvard College. All rights reserved.
Many households pay a marginal price for electricity that exceeds the marginal social cost of supplying that electricity. We show evidence that such pricing schemes can create an incentive to overinvest in energy efficiency. Using hourly smart-meter data for households facing time-invariant increasing block prices, we estimate how air conditioner upgrades affect electricity use. We find that the average participating household reduces consumption by 5%, which provides private savings in the form of lower electricity bills and social cost savings by decreasing generation and pollution costs. The private savings exceed the social savings by an average of 140%, so the average household is faced with an incentive to overinvest in energy efficiency. This incentive to overinvest in energy efficiency would be cut in half if consumers faced any one of three alternative pricing plans with lower marginal price but the same average price.
We provide the first estimates of the potential impact of climate change on cognitive performance and attainment, focusing on the impacts from both short-run weather and long-run climate. Exploiting the longitudinal structure of the NLSY79 and random fluctuations in weather across interviews, we identify the effect of temperature in models with child-specific fixed effects. We find that short-run changes in temperature lead to statistically significant decreases in cognitive performance on math (but not reading) beyond 26°C (78.8°F). In contrast, our long-run analysis, which relies upon long-difference and rich cross-sectional models, reveals an imprecisely estimated effect that is significantly smaller than the short-run relationship between climate and human capital.
We revisit the classic semiparametric problem of inference on a low dimensional parameter θ0 in the presence of high-dimensional nuisance parameters η0. We depart from the classical setting by allowing for η0 to be so high-dimensional that the traditional assumptions, such as Donsker properties, that limit complexity of the parameter space for this object break down. To estimate η0, we consider the use of statistical or machine learning (ML) methods which are particularly well-suited to estimation in modern, very high-dimensional cases. ML methods perform well by employing regularization to reduce variance and trading off regularization bias with overfitting in practice. However, both regularization bias and overfitting in estimating η0 cause a heavy bias in estimators of θ0 that are obtained by naively plugging ML estimators of η0 into estimating equations for θ0. This bias results in the naive estimator failing to be N−1/2 consistent, where N is the sample size. We show that the impact of regularization bias and overfitting on estimation of the parameter of interest θ0 can be removed by using two simple, yet critical, ingredients: (1) using Neyman-orthogonal moments/scores that have reduced sensitivity with respect to nuisance parameters to estimate θ0, and (2) making use of cross-fitting which provides an efficient form of data-splitting. We call the resulting set of methods double or debiased ML (DML). We verify that DML delivers point estimators that concentrate in a N−1/2 -neighborhood of the true parameter values and are approximately unbiased and normally distributed, which allows construction of valid confidence statements. The generic statistical theory of DML is elementary and simultaneously relies on only weak theoretical requirements which will admit the use of a broad array of modern ML methods for estimating the nuisance parameters such as random forests, lasso, ridge, deep neural nets, boosted trees, and various hybrids and ensembles of these methods. We illustrate the general theory by applying it to provide theoretical properties of DML applied to learn the main regression parameter in a partially linear regression model, DML applied to learn the coefficient on an endogenous variable in a partially linear instrumental variables model, DML applied to learn the average treatment effect and the average treatment effect on the treated under unconfoundedness, and DML applied to learn the local average treatment effect in an instrumental variables setting. In addition to these theoretical applications, we also illustrate the use of DML in three empirical examples. This article is protected by copyright. All rights reserved
Machines are increasingly doing "intelligent" things. Face recognition algorithms use a large dataset of photos labeled as having a face or not to estimate a function that predicts the presence y of a face from pixels x. This similarity to econometrics raises questions: How do these new empirical tools fit with what we know? As empirical economists, how can we use them? We present a way of thinking about machine learning that gives it its own place in the econometric toolbox. Machine learning not only provides new tools, it solves a different problem. Specifically, machine learning revolves around the problem of prediction, while many economic applications revolve around parameter estimation. So applying machine learning to economics requires finding relevant tasks. Machine learning algorithms are now technically easy to use: you can download convenient packages in R or Python. This also raises the risk that the algorithms are applied naively or their output is misinterpreted. We hope to make them conceptually easier to use by providing a crisper understanding of how these algorithms work, where they excel, and where they can stumble—and thus where they can be most usefully applied.
This paper provides an ex post evaluation of how changes to a building energy code affect energy consumption. Using residential billing data for electricity and natural gas over 11 years, the analysis is based on comparisons between residences constructed just before and just after a building code change in Florida. While an earlier study using 3 years of data for the same residences showed savings for both electricity and natural gas, new results show an enduring savings for natural gas only. These findings underscore the importance of accounting for all sources of energy consumption when conducting evaluations of building codes. More broadly, the results provide a counterpoint to the growing literature casting doubt on whether ex ante forecasts of energy efficiency policies and investments can provide useful information about actual energy savings. Indeed, more than a decade after Florida's energy code change, the measured energy savings still meets or exceeds the forecasted amount.
In the evaluation of public programs, experimental designs are rare. Researchers instead rely on observational designs. Observational designs that use panel data are widely portrayed as superior to time-series or cross-sectional designs because they provide opportunities to control for observable and unobservable variables correlated with outcomes and exposure to a program. The most popular panel data evaluation designs use linear, fixed-effects estimators with additive individual and time effects. To assess the ability of observational designs to replicate results from experimental designs, scholars use design replications. No such replications have assessed popular, fixed-effects panel data models that exploit repeated observations before and after treatment assignment. We implement such a study using, as a benchmark, results from a randomized environmental program that included effective and ineffective treatments. The popular linear, fixed-effects estimator fails to generate impact estimates or statistical inferences similar to the experimental estimator. Applying common flexible model specifications or trimming procedures also fail to yield accurate estimates or inferences. However, following best practices for selecting a nonexperimental comparison group and combining matching methods with panel data estimators, we replicate the experimental benchmarks. We demonstrate how the combination of panel and matching methods mitigates common concerns about specifying the correct functional form, the nature of treatment effect heterogeneity, and the way in which time enters the model. Our results are consistent with recent claims that design trumps methods in estimating treatment effects and that combining designs is more likely to approximate a randomized controlled trial than applying a single design.
Trustworthy savings calculations are critical to convincing regulators of both the cost-effectiveness of energy efficiency program investments and their ability to defer supply-side capital investments. Today’s methods for measurement and verification (M&V) of energy savings constitute a significant portion of the total costs of energy efficiency programs. They also require time-consuming data acquisition. A spectrum of savings calculation approaches is used, with some relying more heavily on measured data and others relying more heavily on estimated, modeled, or stipulated data.
Machine-learning prediction methods have been extremely productive in applications ranging from medicine to allocating fire and health inspectors in cities. However, there are a number of gaps between making a prediction and making a decision, and underlying assumptions need to be understood in order to optimize data-driven decision-making.
Regulations governing the energy efficiency of new buildings have become a cornerstone of US environmental policy. California enacted the first such codes in 1978 and has tightened them every few years since. I evaluate the resulting energy savings three ways: comparing energy used by houses constructed under different standards, controlling for building and occupant characteristics; examining how energy use varies with outdoor temperatures; and comparing energy used by houses of different vintages in California to that same difference in other states. All three approaches yield estimated energy savings significantly short of those projected when the regulations were enacted.
Measuring consumption and wealth remotely
Nighttime lighting is a rough proxy for economic wealth, and nighttime maps of the world show that many developing countries are sparsely illuminated. Jean et al. combined nighttime maps with high-resolution daytime satellite images (see the Perspective by Blumenstock). With a bit of machine-learning wizardry, the combined images can be converted into accurate estimates of household consumption and assets, both of which are hard to measure in poorer countries. Furthermore, the night- and day-time data are publicly available and nonproprietary.
Science , this issue p. 790 ; see also p. 753
The conventional wisdom for the health care sector is that idiosyncratic features leave little scope for market forces to allocate consumers to higher performance producers. However, we find robust evidence across several different conditions and performance measures that higher quality hospitals have higher market shares and grow more over time. The relationship between performance and allocation is stronger among patients who have greater scope for hospital choice, suggesting that patient demand plays an important role in allocation. Our findings suggest that health care may have more in common with "traditional" sectors subject to market forces than often assumed.
This is an elementary introduction to causal inference in economics written for readers familiar with machine learning methods. The critical step in any causal analysis is estimating the counterfactual-a prediction of what would have happened in the absence of the treatment. The powerful techniques used in machine learning may be useful for developing better estimates of the counterfactual, potentially improving causal inference.
For social scientists, developing an empirical connection between the physical appearance of a city and the behavior and health of its inhabitants has proved challenging due to a lack of data on urban appearance. Can we use computers to quantify urban appearance from street-level imagery? We describe Streetscore: a computer vision algorithm that measures the perceived safety of streetscapes. Using Streetscore to evaluate 19 American cities, we find that the average perceived safety has a strong positive correlation with population density and household income; and the variation in perceived safety has a strong positive correlation with income inequality.
Can open tournaments improve the quality of city services? The proliferation of big data makes it possible to use predictive analytics to better target services like hygiene inspections, but city governments rarely have the in-house talent needed for developing prediction algorithms. Cities could hire consultants, but a cheaper alternative is to crowdsource competence by making data public and offering a reward for the best algorithm. This paper provides a simple model suggesting that open tournaments dominate consulting contracts when cities have a reasonable tolerance for risk and when there is enough labor with low opportunity costs of time. We also illustrate how tournaments can be successful, by reporting on a Boston-based restaurant hygiene prediction tournament that we helped coordinate. The Boston tournament yielded algorithms—at low cost—that proved reasonably accurate when tested “out-of-sample” on hygiene inspections occurring after the algorithms were submitted. We draw upon our experience in working with Boston to provide practical suggestions for governments and other organizations seeking to run prediction tournaments in the future.
Predicting unmeasurable wealth
In developing countries, collecting data on basic economic quantities, such as wealth and income, is costly, time-consuming, and unreliable. Taking advantage of the ubiquity of mobile phones in Rwanda, Blumenstock et al. mapped mobile phone metadata inputs to individual phone subscriber wealth. They applied the model to predict wealth throughout Rwanda and show that the predictions matched well with those from detailed boots-on-the-ground surveys of the population.
Science , this issue p. 1073
This paper evaluates a large-scale appliance replacement program in Mexico that from 2009 to 2012 helped 1.9 million households replace their old refrigerators and air conditioners with energy-efficient models. Using household-level billing records from the universe of Mexican residential customers, we find that refrigerator replacement reduces electricity consumption by 8 percent, about one-quarter of what was predicted by ex ante analyses. Moreover, we find that air conditioning replacement actually increases electricity consumption. Overall, we find that the program is an expensive way to reduce externalities from energy use, reducing carbon dioxide emissions at a program cost of over $500 per ton.
In June 2000, after two years of fairly smooth operation, California's deregulated wholesale electricity market began producing extremely high prices and threats of supply shortages. The upheaval demonstrated dramatically why most current electricity markets are extremely volatile: demand is difficult to forecast and exhibits virtually no price responsiveness, while supply faces strict production constraints and prohibitive storage costs. This structure leads to periods of surplus and of shortage, the latter exacerbated by sellers' ability to exercise market power. Electricity markets can function much more smoothly, however, if they are designed to support price-responsive demand and long-term wholesale contracts for electricity.