Impact of daylight saving time on the Chilean residential consumption

Energy Policy 88 (2016) 456–464

Contents lists available at ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

Impact of daylight saving time on the Chilean residential consumption Humberto Verdejo n, Cristhian Becker, Diego Echiburu, William Escudero, Emiliano Fucks Departamento de Ingenieria Electrica, Universidad de Santiago de Chile, Chile

H I G H L I G H T S


The impact of the application of DST is analyzed in Chilean distribution networks.
The results indicate that there is indeed a marginally small reduction in residential electricity consumption.
A total energy reduction is estimated based on the proposed methodology.

art ic l e i nf o

Keywords: Daylight saving time Econometric model Electrical energy saving Energy consumption

a b s t r a c t Since 1970 Chile has had a Daylight Saving Time (DST) policy in order to reduce residential electricity consumption in the country. The time change was set for the first time by executive decree in 1970, and since that date it was applied every year without great changes until 2010. Since then, and to date, decrees have been set in order to increase the duration of the DST, arguing that there are reasons associated with energy savings that justify the extension of the measure that has been adopted by the authority in recent years. In the present study the impact of the application of DST in terms of decreased household electricity consumption is analyzed using two complementary methods, one based on a heuristic approach and the other using an econometric model. The results indicate that there is indeed a marginally small reduction in residential electricity consumption, although these results are not homogeneous throughout the country. & 2015 Elsevier Ltd. All rights reserved.

1. Introduction The time changing policy in the summer and winter periods, known as Daylight Saving Time (DST), has as its main objective the aim to reduce electricity consumption at the residential level. This is achieved by decreasing the use of artificial light by adjusting the time in such a way that sunlight is used more efficiently, since the change in luminosity at peak demand times in the morning and evening are the events that would produce the largest effect in the use of electricity (Joseph Basconi, 2015; Hill et al., 2010). Last year the main purpose of the DST policy has been questioned in Chile as well as in international experience, mainly due to several factors of variable and diverse nature, such as weather changes, energy efficiency policies aimed at reducing the consumption of electricity for artificial lighting in homes and n

Corresponding author. E-mail addresses: [email protected] (H. Verdejo), [email protected] (C. Becker), [email protected] (D. Echiburu), [email protected] (W. Escudero), [email protected] (E. Fucks). http://dx.doi.org/10.1016/j.enpol.2015.10.051 0301-4215/& 2015 Elsevier Ltd. All rights reserved.

companies, the use of air conditioning in summer and heating in winter, geography and morning demand peak. In particular, international experience has shown that the effect of applying DST is at least questionable, because some studies show that the effect is practically nil (Krarti and Hajiah, 2011; Momani et al., 2009; Kellogg and Wolff, 2008). Cases have also been reported in which the effect has been negative, i.e., applying the DST has increased the consumption of electricity instead of reducing it, which is its purpose (Kotchen and Grant (2011)). Considering these arguments, it is reasonable to question whether the application of this policy is still a valid alternative with the purpose of reducing the consumption of electricity. Most of the studies concerning the impact of the time change in the countries do it from an energy approach, however, it is important to consider that the effects of applying such measures are not only limited to the purely energy, but also to the social live, psychological, financial and environmental effects. The structure of the paper is the following: Section 2 describes the methodology used to evaluate the impact of the Daylight Saving Time policy, following an heuristic approach and a

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

econometric model (Differences in Differences); Section 3 refers to the kinds of data used in the evaluation of the DST; Section 4 gives the results obtained by applying the proposed methodology to four regions of Chile; and Section 5 shows the Conclusion and Policy Implications of the study.

2. Methodology For the development of the present study residential feeders representative of four cities of Chile: Arica, Santiago, Concepción and Punta Arenas, were analyzed. The regions considered for the analysis were defined by the Ministry of Energy of Chile, which also provided the consumption data associated with each of the feeders. The information on residential consumption for carrying out the study was selected based on the dates on which the corresponding time changes took place in summer (ST) and winter (WT) (see Fig. 1); the green bands indicate 10 workdays in both directions with respect of the time change, and the red line 23 workdays under the same conditions. This information in a long range of time is not precise at all, since the systems of Supervisory Control And Data Acquisition (SCADA), responsible to get observations about residential consumption, fail in the most of time in their communications. In 2013 WT started on April 27 and ST on September 7, while in 2014 WT started on April 26 and ended on September 6. Since in the literature there is no empirical research on the application of the DST policy in Chile, in a large part of this study methodological strategies gathered from the observation of international experience were applied, with slight variations and shades in accordance with the Chilean context. Following this logic, the method described to select the data tries to be as impartial as possible, so that it can approach what is known as simple random sampling (Gujarati and Porter, 2011). The aim underlying the selection of the two time strips mentioned in Fig. 1 is due to the fact that in Chile the DST policy is applied uniformly to the whole territory at two times in the year, on the dates mentioned, and to the physical restrictions for access to information belonging to the country. In agreement with what was reported in U.D. of Energy (2008), it was decided to analyze two time strips around the two times of the year on which the time change took place, with data for two years when the four changes took place on different dates. The above is associated with the fact that the windows analyzed every year would not be exactly the same, i.e., they would contain different days and dates. We would therefore have two crossing sections merged in time. This allows analyzing the effect of DST considering weather factors (temperature) or intrinsic to each time window (days, months, etc.), which may have an effect on

April

September

27

7

2013

26

6

2014

WT

ST

January

Fig. 1. Selected time intervals. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

457

consumption after every time change. In this way we have a wider perspective that takes into account a greater range of determining elements at the time of giving an explanation of the performance of consumption. This would allow having a more detailed description than that which can be obtained from the traditional statistical-descriptive methods. The following section details the general methodology applied to the city of Santiago, which will then be extended to the cities involved in the evaluation. 2.1. Heuristic method in Santiago The heuristic method was used as a way of approaching the saving in electric energy after the application of the DST, which is done by comparing the electric demand between two periods of time in which the time change is applied, without using a formal mathematical modelling. The approach is heuristic in the sense that through observation over a given range of days around the time change of the average electricity demand profiles observed in the days preceding and following the time change, a comparison is made trying to approach theoretically the gross effect of the policy (U.D. of Energy, 2008). The methodology for the application of the heuristic method considers the following process: 1. First, the geographical area where the impact of DST is studied must be selected. 2. Then, the information of hourly electric power demand from a distribution company present in the region should be obtained. 3. The exact moment when the time change occurs is determined. Following that, two time periods are chosen, one of them prior to the date where the time change occurs and another where the change has already been applied. For example, in Chile, during the 2014, two moments were set in which the time change was applied, associated with four analysis periods. (a) End of Summer Time: April 26, 2014. Since that day, a period corresponding to the summer time (ST) and one after the change (winter time, WT) was established. (b) Start of Summer Time: September 6, 2014. On this day, a period prior to the time change was established (during WT) and one after the ST was also chosen. 4. It is necessary that the chosen periods are similar in number of days, such that the sample or data window is calculated with the same number of observations for each period, thereby avoiding biases in the trends. For this purpose, it should be noted that: (a) Only weekdays are considered for the Heuristic model, because the behaviour of electric consumption on Saturday and Sunday are considerably different from weekdays. (b) Similarly, the holidays are omitted from the analysis. 5. Once the hourly electrical demand curves for each day of the periods previously established were obtained, an equivalent curve was obtained for each analyzed time interval. This was achieved by averaging the value of the electrical demand, independently for each daylight hours. This allowed us to obtain an average demand curve per hour for each period, both before and after the time change. As an example, in each one of the curves provided in Fig. 2, the methodology explained above is applied. 6. After both equivalent curves of electricity demand (associated with each time change) were obtained, the time intervals considered as treated hours were determined, that is, to determine the time interval where the demand for electrical power changes due to the application of the policy of time change; that can be seen in a graphical way looking for the variations in the curve. For example, in Fig. 2, the time interval corresponds to the hours between 5:00 and 9:00 am and 17:00 and 23:00 in the evening.

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

Once the heuristic methodology is applied to the available information, hourly demand curves like those described in Fig. 2 can be obtained. This information is vital from the descriptive standpoint, because it allows to state the following:


First, there is a clear difference between the consumption in the




summer time compared to that of winter; this is evidenced basically because the summer consumption curve is displaced to the right with respect to the consumption curve during the winter time. This may be interpreted, at first sight, as an indication that there are effects in summer that make consumption be displaced in time. Second, it is found that the peak values of consumption occur in the morning and in the evening, also having the minimum values of the curves in the early morning, see Fig. 2. In the period between ten in the morning and five in the afternoon the variations seen are small. This agrees with the reports in the international literature, see (Sarwar et al., 2010), where it is stated that the DST policy will only have an effect in the times of day related to dawn and dusk. This last point is of vital importance in the methodological strategy for evaluating the impact of the DST policy, and therefore be able to apply the evaluation to the different regions of Chile that have different geographic locations.

2.2. Econometric method in Santiago The impact assessment methodology used in this paper is known in the empirical literature as Differences in Differences (Wooldridge, 2010). It has as objective, to calculate the difference recorded in a group of affected or treated hours by de DST policy, in other words, those hours that have suffered a significant alteration of light, and where the latter is relevant. The previous in the sense that on average, it alters consumption patterns 1 Each ratio is related to the division between the demand for the period where DST is applied and when it is not applied. This division is performed point by point, in this case, for each hour.

10

Electricity Demand (MW)

7. Once the treated hours were defined, the division (ratio1) of these demand curves was conducted in order to obtain the impact that the time change has in electricity consumption, as it is shown in Fig. 3. After realizing this division, two types of results can be established beforehand: (a) In the event that any difference in the demand for electricity exists due to the time change, this would be reflected in the form of irregularities in the curve obtained after making the division. (b) Otherwise, if there is no significant effect on the demand for electricity, the obtained curve will show a flat pattern, where the abrupt variations due to the time change will be minimal. This case indicates that a shift in the demand curve does not exist in that time interval, and therefore, the pattern of electricity consumption by customers would not be affected by the application of a DST policy. 8. With the ratio obtained, we proceed to make a linear regression on the intervals defined as the treated hours, in order to reflect the pattern of demand that would have happened if the time change had not been applied. These new points define amplitudes regard to the points of the original demand curve, as in the case of being negative, shows energy savings. 9. Finally, using the explanation shown above, for both time changes produced in each year, it is possible to obtain an indicator of the savings or increase in annual electricity consumption obtained from the application of the DST policy.

WT ST 8

6

4

2 0

5

10 15 Hours

20

First quarter.

10

Electricity Demand (MW)

458

WT ST 8

6

4

2 0

5

10 15 Hours

20

Second quarter. Fig. 2. Hourly demand in Santiago for the year 2013. (a) First quarter. (b) Second quarter.

substantially at these hours, known as morning and evening peaks. Regarding the group of hours not affected by the policy, because in them, the light has little alteration as to affect consumption patterns (known as the hours in the control group). In addition, observations of these two groups are collected (treated and untreated), both before and after applying the DST policy. This information is used to build an average difference of initial energy consumption before the implementations of the policy, between the two groups. Then, once the policy is uniformly applied, this differential is recalculated with the hope that the policy has generated a significant effect on the groups treated. Thus, when comparing (difference) these two mean differences, it is obtained a measure (second difference or difference in difference), useful as proxy for the average treatment effect, in other words, the impact of having applied the DST policy on a representative hour of the analyzed peak time (morning /evening).

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

Therefore, the results obtained only have internal validity (Stock and Watson, 2012). It is important to mention that the methodology used here is ad hoc, indicating that it can only be extrapolated to geographical locations where the DST is uniformly applied every hour of the day, throughout the geographic national territory having a determined seasonal duration, as in Chile. If it is intended to assess the impact or simulate the impact of applying it on any other DST schema, it would be necessary to use a different methodological framework to the one developed in the present study. Keeping in mind that in this case there is no control group because the DST policy is applied uniformly to the whole territory, it is not possible to know what would have happened with the energy consumption at peak times on the summer time days by remaining on the winter time regime (  4:00 UTC), i.e., we do not have what from now on will be called a counterfactual. To that end the results delivered by the heuristic model were used, together with basic descriptive statistical tools to observe and in this way create artificial counterfactuals. These elements allow specifying an econometric model that was estimated by the method of differences in differences (DD) (Wooldridge, 2010). Looking at the comparison made by the heuristic method between the hourly consumption curve before and after applying the DST policy (Fig. 2), there are time intervals in which the summer time curve is transferred marginally. At some points neighbourhoods the two curves even coincide. This confirms the fact that we can divide the day into various time ranges or intervals: some where the policy affects consumption (treated times) and others where the policy has no effect, or its influence is marginal (control times, or counterfactuals). For Santiago the following time intervals are defined:

459

Table 2 Descriptive statistics time ranges with DST. Variable

Obsa

Average

Std. Dev.b

Min

Max

consug~DST consum̃ DST consud̃ DST ̃ consutDST

1114 1115 1566 1553

12.05799 10.90870 13.57005 18.01842

3.767201 3.116854 2.768320 5.214768

6.0 5.4 7.6 7.4

25.4 19.5 26.6 31.7

a b

Observation. Standard deviation.

the times that will be analyzed as treated or affected. However, in the just mentioned variations there is a number of observable as well as subjacent factors that are not taken into account but may affect the analysis that is being made of the DST policy. To evaluate the effect of each of the possible variables that affect consumption at different times and have not been considered in the descriptive analysis, an econometric model that can help to approach causal effects is specified. Moreover, it will be possible to distinguish not only between times observed before or after the time change, but also between a time group called control or counterfactual, and two different groups of treated times (which will be the peak times in the models). The impact of the DST policy for the city of Santiago is analyzed by estimating the following populational regression function (1):

Ln (Consumpt )h = δ 0 + δ1Peakmh + δ2 Peakth + δ3 ST (DST )h + δ4 Temph + δ5 CDSTh


From 00:00 to 04:59 it is called early morning control ( consumo _madrug ), considered as a control time interval.


From 05:00 to 09:59 it is called morning peak (consumo_peakm), considered as a treated time interval.


From 10:00 to 16:59 it is called midday ( consumo_mid ), con-

+ δ6 Holidh + δ7 Sath + δ8 Sundh + δ 9 y2014h + ζcfmidh

sidered as a control interval.


From 17:00 to 23:59 it is called evening peak ( consumo_peakt ), considered as a treated time interval. Upon dividing the day into these four time categories, Tables 1 and 2 show the descriptive consumption statistics in the corresponding time blocks, arranged as they were described above. Let us consider the consumption in each time range: the days on which there was no application of the time change policy are denoted as No DST, and those periods in which there is a time change policy as DST. In this respect, it should be mentioned that the variation of the measures registered at all times except the evening peak shows a decrease of approximately [1.9–2.0] MWh, while in the evening peak it decreases by approximately 5 MWh. It is therefore seen that without any control of some variable, the evening peak time shows the largest variation between one period and another. Furthermore, this analysis shows that the times that will be part of the policy's control group do not differ much from Table 1 Descriptive statistics time ranges without DST. Variable

Obs

Average

Std. Dev.

Min

Max

consug~NODST consum̃ NODST consud̃ NODST ̃ consutNODST

641 640 891 886

13.94587 12.82984 15.53199 23.01117

4.298516 3.653215 3.061427 5.836728

7.2 6.6 7.4 9.0

28.1 22.2 26.8 35.8

+ β1 (PeakmST (DST ))h + β2 (PeakST (DST ))h + θ ⁎DuHOURh + α ⁎DuMONTHh + μh

(1)

The variable Ln (Consumpt )h represents the natural logarithm of consumption in MWh, observed in the Rojas Magallanes feeder at time h of our sample. That feeder was specified as representative of consumption in Santiago because it is located in a residential area that has a substantial part of the population of Santiago. The dichotomous variable Peakmh assigns a value of one to those times that belong to what we called morning peak, otherwise it is zero. The dichotomous variable Peakth assigns a value of one to those times that belong to what we called evening peak, otherwise it is zero. The dichotomous variable cfmidh assigns a value of one to those times that belong to what we called midday counterfactual, otherwise it is zero. This variable allows modelling two control groups when midday and midnight are used as different groups. The dichotomous variable ST (DST )h assigns a value of one to those observations that belong to the summer time, and therefore they are observations on which the DST is being applied. It should be noted that this variable does not measure any effect of the policy, but by adding it, an extra coefficient is included in the

460

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

analysis that allows controlling those factors that are inherent to the category of being the time in the summer time, and they may not have a direct relation with the DST policy itself. The continuous variable Temph uses the temperature in Santiago, recorded at the weather station of Pudahuel and provided by the Direccion Meteorologica de Chile. The control variable CDSTh uses days in which the policy cannot have an effect, prestart of school activities (March 1 to 10) and national holidays (the week of September 18). Its value is one on those dates and zero otherwise. The dichotomous variables Holidh, Sath and Sundh assign values of one to those times that belong to some day that is a holiday, Saturday or Sunday, and zero otherwise. The dichotomous variable y2014h assigns a value of one to those times that were recorded in the year 2014. This variable allows getting the effects associated only with the change of year, such as the country's economic growth. The interaction terms PeakmST (DST )h and PeaktST (DST )h arise from multiplying peakmh and peakth, respectively, by ST (DST )h , the purpose of measuring the impacts of the DST policy on the morning and evening peak times. The dummy variables vector DuHOURh exerts a control for each hour of the day (24 categories, reference time 00:00). It should be mentioned that in the econometric estimation of each model, to avoid multicolinearity problems between the variables, no dummy variable corresponding to the group used as control was added. The dummie variables vector DuMONTHh controls for each of the months that are being analyzed (there are six categories, with the month of March as reference). The criteria by means of which it was chosen to control by days, hours, or months agree with the practice applied in studies reported in the international literature. It is important to point out that within the parameters designated with Greek letters, the betas were specified for the variables that considered the difference in differences estimator for each treated group. Also, the parameters of the last two dummy variables of (1), which obey other possible controls, were highlighted with a superscript asterisk (*). 2.3. Methodological differences in the regions Once the efficacy of the model used for the case of the nonexperimental data of Santiago (Rojas Magallanes Feeder) had been proved, the methodology used here to approach a quantitative evaluation of the impact of DST policy was replicated in a restricted way in each of the regional capitals considered in the present study, i.e., Arica, Concepción and Punta Arenas. When it is said that the model used for the case of Santiago was applied in a restricted way, it means that in each region its functional form was altered (specifically the control and treatment variables, with their corresponding interaction terms), with the purpose of flexibilizing the original estimation model by difference in differences. This allows the representation of the specific characteristics shown by the hourly residential energy consumption in each of the geographic sectors analyzed (see Appendix B). The heuristic method for the regions generates consumption curves for the summer time and the winter time with the purpose of getting the corresponding artificial counterfactuals (Departamente de Ingenier Elctrica, 2015). Table 3 shows an example for the case of Arica in the first quarter of 2013, where the interval of the control hours is highlighted in bold. Once this procedure has been applied, the information needed to specify a different model (equation) for each city is available (see Appendix A).

Table 3 Example for the case of Arica, first quarter. Time

ST Consumption (Wh)

WT Consumption (Wh)

Ratio

VAR

00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00

482.40 453.60 419.04 378.72 368.64 355.68 370.08 447.84 446.40 590.40 616.32 617.76 624.96 619.20 577.44 616.32 642.24 604.80 534.24 463.68 545.76 557.28 551.52 555.84

496.80 498.24 502.56 476.64 455.04 442.08 436.32 446.40 452.16 535.68 629.28 614.88 617.76 573.12 558.72 580.32 591.84 584.64 540.00 544.32 586.08 567.36 564.48 560.16

0.971 0.910 0.834 0.795 0.810 0.805 0.848 1.003 0.987 1.102 0.979 1.005 1.012 1.080 1.034 1.062 1.085 1.034 0.989 0.852 0.931 0.982 0.977 0.992

0.0290 0.0900 0.1660 0.2050 0.1900 0.1950 0.1520  0.003 0.0130  0.102 0.0210  0.005  0.012  0.080  0.034  0.062  0.085  0.034 0.0110 0.1480 0.0690 0.0180 0.0230 0.0080

3. Data used Due to the particular geography of Chile (a long and narrow country), it is difficult to find a region that is equivalent to and representative of the whole country. It should be noted that each region presents important differences at the time of making an analysis of the DST policy in the whole territory, i.e., if there are variations due to weather conditions, number of hours of daylight according to latitude, prevailing economic activity in the region, percentage of residential consumers, etc. Considering the above, and keeping in mind what was pointed out at the beginning of the present study, it was decided to analyze four cities in order to determine the impact of the DST. The cities of Santiago and Concepción were chosen because they have the largest population in the country. The cities of Arica and Punta Arenas are located geographically at the country's far north and south ends. Both the econometric and the heuristic models require the electricity demands of characteristic residential feeders (or else the consumptions of electricity) for each city, which were obtained from the information gathered by the Ministerio de Energí a de Chile and provided by the utilities that deliver electrical energy to the residential customers of the regions chosen in this study: Chilectra (Santiago), EMELARI (Arica), CGE Distribución (Concepción) and EDELMAG (Punta Arenas). The information associated with temperature was provided by the Dirección Meteorológica de Chile. This information considers the temperature for each hour of the day for the regions and the study period. With respect to the residential load component of the total electricity consumption corresponding to each city, it was obtained from the Instituto Nacional de Estadísticas (INE, 2015), and the BT1 prices are available in the web sites of each of the electrical utilities. 4. Results Once the treatment and control groups corresponding to each city were defined, the adjusted regression polynomial was

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

Table 4 Summary of the results of the econometric model.

461

2

Santiago

Arica

Concepción

P.A

Morning Peaka Evening Peak C.F.Mb STc (DST) Temperature DST controls Holidays Saturdays Sundays Year 2014 PDSTd (morninga ) PDST (eveninga) C.H.De C.M.Af

52.92% 21.20% N/A  2.11%  2.16%  1.90%  7.38%  1.39% 1.38%  27.94%  4.40%  7.76% Effect (  ) Effect 

 4.00% N/A N/A  0.90% 1.06%  3.80%  23.65%  7.97%  32.97% 3.08%n  1.46% N/A Effect ( þ /  ) N.Sn

 33.60% 33.21% 4.90%  9.28%  0.78%  1.70%  2.89%  0.97%  0.52%n 4.06% 16.07% 1.24%n Effect (  ) Effect ( þ)

N/A 46.06% N/A  4.40%  0.70% 7.30%  6.28%n  6.97%  3.48% 26.94% N/A  12.96% Effect ( þ/  ) Effect ( þ)

a

Case that applies only to Santiago and Concepción. Midday counterfactual. c Summer time. d DST peak. e Controls per hours of the day. f Controls per analyzed months. n Not significant variables. b

estimated according to the energy consumption reality of each geographic sector analyzed (see Appendix A). The estimation of the corresponding difference in differences models was made using the STATA S.E statistical software. The results are shown in Table 4. From the estimation made we get that all the proposed models are significant at the populational level, and there are sufficient degrees of confidence to ensure the statistical validity of the results. On the other hand, the models proposed for each case, except Arica, account for about 70% of the variations registered in the logarithm of the consumption. This allows arguing that the set of variables incorporated (which except for the control groups were the same for all the geographic applications) help explain a substantial percentage of the variations of the consumption logarithm. The 36% of explanation achieved in Arica brings up questions with respect to the energy consumption situation in the north of Chile, which seems to follow different dynamics from the rest of the regions analyzed. In this respect, the results shown in Table 4 show that when the temperature variable increases one degree, in general the energy consumption decreases somewhat less than 1% per hour (an effect that is doubled in the case of Santiago), while for the case of Arica the same temperature change causes an energy consumption increase of 1%. This indicates that the differences related to the weather (and its effects on energy consumption) will not be taken into account by simply incorporating the temperature variable, but rather, as known in geography, there is a large number of variables that can be mentioned at the time of explaining the weather of a city. The study gives empirical validity to the hypothesis that on Saturdays, Sundays and holidays energy consumption decreases. However, the observation that this does not happen in Santiago on Sundays poses interesting research questions in terms of public policy. When the policy was applied to only one group of treated hours, it gave a result that is unequivocally negative. Applying a Daylight Saving Time policy decreases the electrical consumption, and this effect is significant at the individual level. However, when the specification of the model was aimed at defining more than one treated group, the negative direction of the effect of the policy remains in Santiago, but it turns positive in Concepción, a city in which the effect on the evening peak with respect to the morning peak is not significant (see Fig. B2). The difference in the results

Ratio Interpolation

1.8

Ratio WT/ST 2014

Variable

1.6 1.4 1.2 1 0

5

10 15 Hours

20

Fig. 3. Ratio from the first quarter, 2014, for a 23 days window.

obtained in Concepción can be justified by the fact that, since the region is located near the coast, it is forced to start port activities earlier than the remaining cities that were studied. As mentioned previously, the variables that use the actual effects of the policy correspond to the interactions of the morning and evening peaks with the DST. The results related to these variables in each city are shown in Table 4, where it is seen that in the case of Arica it does not apply during the evening because it does not have an appreciable peak during that time (Fig. B1), while in Punta Arenas the hourly consumption curve is altered only in the evening (Fig. B3). Each factual has a time period determined by the observation of the time consumption curves (see Appendix B). Having this clear, it is possible to determine the daily percentage due to the application of the policy. For example, in Santiago

Savings%day =

7.76%· 7 h + 4.40%· 5 h = 3.18% 24

In 2014 the winter time was between April 27 and September 7 (see Fig. 1), for a total of 133 days. Since this was the official time in Chile up to 2014 ( −0 4:00 h UTC), the days affected by DST are those on which the summer time is applied ( −0 3:00 h), so the percent reduction in the whole year is defined by

Savings% year = 3.18%·

365 − 133 = 2.02% 365

However, this value corresponds only to the residential sector of Santiago (27% of the citys total load) (INE, 2015), so this finally defines an annual electric energy saving of 0.55%. Applying the same development, the annual energy saving results for the cities considered in this study are shown in Fig. 4. Knowing the consumption of electricity in each city (INE, 2015) and the BT1 rates corresponding to each distributor of electricity in the zone, Table 5 shows a summary of the annual energy and monetary savings in the corresponding cities in Chile.

5. Conclusion and policy implications The results obtained from the study allow us to infer that the application of a daylight saving time policy shows an overall saving of electrical energy in the residential customers from Chile. This overall saving is explained by the following: an additional result of this study shows that the average residential consumption

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

Annual Savings of Electrical Energy %

462

0.6

0.55 0.48

0.4 0.2 4 · 10−2 0 −0.2 −0.32

−0.4 Arica

Santiago Concepcion

P.A

Fig. 4. Annual percent electric energy savings.

was reduced in a 3.18% as a result of the application of summer time. The above is obtained considering a consumption reduction of 4.4% in the morning peak hours (5–10 am), and 7.76% in the evening peak hours (17–23 h). The annualized average consumption reduction as a result of the existence of summer time all year corresponds to 2.02%. The reduction of 3.18% is obtained assuming a constant consumption during 24 h per day. The average reduction can be extrapolated to Santiago, giving us a reduction of 0.5% of total demand in Santiago. Meanwhile, for the other regions different from Santiago, the saving is greater in Punta Arenas than in Arica, while in Concepcin the policy seems to have an opposite effect to what was anticipated/expected. It is shown that in Punta Arenas, considering only one representative hour of the afternoon peak, the DTS policy generates an average decrease of 28.13%, as compared to an average of 46.06% increase in the consumption in some of these hours. On the other hand, the same average in Arica reinforces the fact that critical hours, recognized as the morning peak have lower average consumptions when compared to the other hours, that is to say, in Arica the policy has the expected result of a further reduction in the consumption during critical hours. In relative terms, the only significant effect found in Concepcin is that the DST policy increases the consumption in a representative hour of the morning peak in a 47.8%. In simple terms, the DST policy is very effective in Santiago, Arica and Punta Arenas (in order of importance), however, it is very ineffective in Concepcin. Another important evidence this study presents which is of paramount importance when designing energetic policies is the effect of geographic heterogeneity. An effective form to exemplify this idea, particularly in Chile, is by analyzing the effects of the temperature variable on the energy consumption, which seems to

have the opposite effect in the cities of Punta Arenas and Concepcin (less than a 1%), whereas it has an important effect (2.16% downwards trend) in Santiago. Conversely, in Arica, the temperature effect is positive. We believe that this phenomena can be the result of the characteristic dry climate of this particular city, as well as the fact that it is closer to the Equator line. Regarding to residential customers with hourly rate, one might assume that they behave similarly to Rojas Magallanes. In that case, it is possible to argue that these customers, during the peak evening rate (from 18 to 22 h), with a 30% surcharge on the corresponding rate, will be able to save approximately 7.76% of energy during the most expensive hours. Given the results in each city, the DST energy policy is strongly dependent on the geographical location. Chile is a long and narrow country, where its population is mostly situated between meridians 70 and 75. Arica, Santiago and Punta Arenas are cities near the meridian 70, while Concepción is located between 70 and 75. This geographic feature is linked directly with the sunrise and sunset, which justifies the increased residential consumption produced at the peak of the morning in the city of Concepción and the reduction produced in the peak of the evening in all cities. Finally, in relation to MWh, the saving in Santiago is significantly more important than is Punta Arenas and Arica. These are the three regions where the policy has had negative effects in the consumption and positive ones concerning energy saving. The positive effects found in Concepcin, as opposed to what is reported in the specialized literature, generate an energy deficit as a result of the application of the policy. It should be mentioned that it would be necessary to consider other aspects not related to economic concerns, it is of upmost importance to consider other determinant aspects in the behavior of the population, such as psychological effects, public transportation and traffic accidents, social and biological subjects, when the government decided to make a change in the public policy, concerning the daylight saving time application and a possible change in the Chilean official time zone.

Acknowledgments The financial support of this study by the Division de Seguridad y Mercado Eléctrico, Ministerio de Energía de Chile, in particular to María José Reveco and by Fondecyt under Project 11130169 (Comisión Nacional de Investigación Científica yTecnológica de Chile) is gratefully acknowledged.

Appendix A. Econometric model for regions A.1. Arica

Table 5 Annual energy and monetary savings. City

Distributor

Savings (GWh)

Savings MM US$a

Arica Santiago Concepción P.Ab

EMELARI CHILECTRA CGE Distribucin EDELMAG

0.97100 94.3120  21.322 1.88800

0.2121 15.312  3.795 0.3635

a b

Value of the dollar on Thursday, March 5, 2015, equivalent to ch$622. Punta Arenas.

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

463

+ δ3 Temph + δ4 CDSTh + δ5 Holidh + δ6 Sath + δ7 Sundh + δ8 y2014h + β1 (PeakST (DST ))h + θ ⁎DuHOURh + α ⁎DuMONTHh + μh

(A.1)

Electricity Consumption (MWh)

Ln (Consumpt )h = δ 0 + δ1Peakmh + δ2 ST (DST )

WT ST 600

500

400

300 A.2. Concepción

0

5

10 15 Hours

20

Fig. B1. Hourly consumption in Arica second quarter 2014.

Ln (Consumpt )h = δ 0 + δ1Peakmh + δ2 Peakth

Electricity Consumption (MWh)

+ δ3 ST (DST )h + δ4 Temph + δ5 CDSTh + δ6 Holidh + δ7 Sath + δ8 Sundh + δ 9 y2014h + ζcfmidh + β1 (PeakmST (DST ))h + β2 (PeakST (DST ))h

WT ST

1.4 1.2 1 0.8 0.6 0.4

+ θ ⁎DuHOURh + α ⁎DuMONTHh + μh

(A.2)

0

5

10 15 Hours

20

Fig. B2. Hourly consumption in Concepcion second quarter 2014.

A.3. Punta arenas

+ δ3 Temph + δ4 CDSTh + δ5 Holidh + δ6 Sath + δ7 Sundh + δ8 y2014h + β1 (PeakmST (DST ))h + θ ⁎DuHOURh + α ⁎DuMONTHh + μh

(A.3)

Electricity Consumption (MWh)

1.2 Ln (Consumpt )h = δ 0 + δ1Peakh + δ2 ST (DST )h

WT ST 1 0.8 0.6 0.4

0 Appendix B. Hourly consumption in other cities See Figs. B1–B3.

5

10 15 Hours

20

Fig. B3. Hourly consumption in Punta Arenas second quarter 2014.

464

H. Verdejo et al. / Energy Policy 88 (2016) 456–464

References Joseph Basconi, 2015. The impact of daylight savings time on electricity consumption in indiana, Tech. rep., University of Notre Dame, Department of Chemical and Biomolecular Engineering. Hill, S., Desobry, F., Garnsey, E., Chong, Y.-F., 2010. The impact on energy consumption of daylight saving clock changes. Energy Policy 38 (9), 4955–4965. Krarti, M., Hajiah, A., 2011. Analysis of impact of daylight time savings on energy use of buildings in kuwait. Energy Policy 39 (5), 2319–2329. Momani, M.A., Yatim, B., Ali, M.A.M., 2009. The impact of the daylight saving time on electricity consumptiona case study from jordan. Energy Policy 37 (5), 2042–2051. Kellogg, R., Wolff, H., 2008. Daylight time and energy: Evidence from an australian experiment. Journal of Environmental Economics and Management 56 (3), 207–220. Kotchen, M.J., Grant, L.E., 2011. Does daylight saving time save energy? evidence from a natural experiment in indiana. Review of Economics and Statistics 93

(4), 1172–1185. Gujarati, D.N., Porter, D.C., 2011. Econometria Básica-5. McGraw Hill Brasil. U.D. of Energy, 2008, Impact of extended daylight saving time on national energy consumption, Tech. rep., U.S. Department of Energy. Sarwar, R., Chakrabartty, R., Ahmed, N., Ahmed, K., Ahsan, Q., 2010. Effect of daylight saving time on bangladesh power system, in: Electrical and Computer Engineering (ICECE), 2010 International Conference on, IEEE. pp. 291–293. Wooldridge, J.M., 2010. Econometric analysis of cross section and panel data. MIT press. Stock, J.H., Watson, M.W., 2012. Introduction to Econometrics: Global Edition. Pearson Education. U.d.S.d.C. Departamente de Ingenier Elctrica, Cambio de horario y su efecto en el consumo de energa elctrica, desarrollado para el Ministerio de Energa de Chile, p. 49, 2015. INE, 2015, http://www.ine.cl/canales/chile_estadistico/estadisticas_economicas/ energia/series_estadisticas/series_estadisticas.php.