Temperature correction of energy consumption time series Sumit Rahman, Methodology Advisory Service, Office for National Statistics 8 08 l-0 Ju 7 07 l-0 n- Ja Ju 6 06 l-0 n- Ja Ju 5 05 l-0 n- Ja Ju 4 04 l-0 n- Ja Ju 3 03 l-0 n- Ja Ju 2 02 l-0 n- Ja Ju 1 01 l-0 n- Ja Ju 0 00 l-0 n- Ja Ju 9 99 l-9 n- Ja Ju 8 98 l-9 n- Ja Ju 7 97 l-9 n- Ja Ju 6 96 5 l-9 n- Ja Ju n- Ja 95 l-9 Ju n- Ja Gigawatt hours Final consumption of energy – natural gas • Energy consumption depends strongly on air temperature – so it is seasonal Gas consumption 120000 100000 80000 60000 40000 20000 0 Average monthly temperatures • But temperatures do not exhibit perfect seasonality deviations in temperature from long-term monthly averages +4.0 +2.0 +1.0 +0.0 -1.0 -2.0 -3.0 Jan-10 Jan-09 Jan-08 Jan-07 Jan-06 Jan-05 Jan-04 Jan-03 Jan-02 Jan-01 Jan-00 Jan-99 Jan-98 Jan-97 Jan-96 Jan-95 Jan-94 Jan-93 Jan-92 -4.0 Jan-91 deviation (degrees Celsius) +3.0 Seasonal adjustment in X12-ARIMA • • • • Y=C+S+I Series = trend + seasonal + irregular Use moving averages to estimate trend Then use moving averages on the S + I for each month separately to estimate S for each month • Repeat two more times to settle on estimates for C and S; I is what remains Seasonal adjustment in X12-ARIMA • Y=C×S×I • Common for economic series to be modelled using the multiplicative decomposition, so seasonal effects are factors (e.g. “in January the seasonal effect is to add 15% to the trend value, rather than to add £3.2 million”) • logY = logC + logS + logI Temperature correction – coal • In April 2009 the temperature deviation was 1.8°(celsius) • The coal correction factor is 2.1% per degree • So we correct the April 2009 consumption figure by 1.8 × 2.1 = 3.7% • That is, we increase the consumption by 3.7%, because consumption was understated during a warmer than average April 8 08 l-0 Ju 7 l-0 n- Ja 07 unadjusted Ju 6 06 l-0 n- Ja Ju 5 l-0 n- Ja 05 thousands of tonnes 10000 Ju 4 04 l-0 n- Ja Ju 3 03 l-0 n- Ja Ju 2 02 l-0 n- Ja Ju 1 01 0 l-0 n- Ja Ju n- Ja 00 l-0 Ju 9 99 l-9 n- Ja Ju 8 98 l-9 n- Ja Ju 7 97 l-9 n- Ja Ju 6 96 l-9 n- Ja Ju 5 95 l-9 n- Ja Ju n- Ja Current method – its effect Coal consumption seasonally adjusted 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 n- 8 08 l-0 Ju 7 07 l-0 n- Ja Ju 6 06 l-0 n- Ja Ju 5 l-0 n- Ja 05 thousands of tonnes Coal consumption Ju 4 04 l-0 n- Ja Ju 3 03 l-0 n- Ja Ju 2 02 l-0 n- Ja Ju 1 01 l-0 n- Ja Ju 0 00 l-0 n- Ja Ju 9 99 l-9 n- Ja Ju 8 98 l-9 n- Ja Ju 7 97 l-9 n- Ja Ju 6 96 l-9 n- Ja Ju 5 95 l-9 n- Ja Ju Ja Current method – its effect seasonally adjusted 8000 temperature corrected and seasonally adjusted 7000 6000 5000 4000 3000 2000 1000 0 Regression in X12-ARIMA • Use xit as explanatory variables (temperature deviation in month t, which is an i-month) • 12 variables required • In any given month, 11 will be zero and the twelfth equal to the temperature deviation Regression in X12-ARIMA • Why won’t the following work? 12 log Yt i xit log Ct St I t i 1 Regression in X12-ARIMA • So we use this: 12 log Yt i xit ARIMA i 1 Regression in X12-ARIMA • More formally, in a common notation for ARIMA time series work: 12 ( B)( B )(1 B) (1 B ) (log Yt i xit ) 12 d 12 D i 1 ( B)( B ) t 12 • εt is ‘white noise’: uncorrelated errors with zero mean and identical variances Regression in X12-ARIMA • An iterative generalised least squares algorithm fits the model using exact maximum likelihood • By fitting an ARIMA model the software can fore- and backcast, and we can fit our linear regression and produce (asymptotic) standard errors Coal – estimated coefficients coefficient (percentage) 20 15 10 5 0 -5 -10 -15 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Interpreting the coefficients • For January the coefficient is -0.044 • The corrected value for X12 is log Yt • The temperature correction is e 12 x i 1 i xit • If the temperature deviation in a January is 0.5°, the correction is e ( 0.0440.5) 1.022 • We adjust the raw temperature up by 2.2% • Note the signs! i it Interpreting the coefficients • If i xit is small then e i xit 1 i xit • So a negative coefficient is interpretable as a temperature correction factor as currently used by DECC • Remember: a positive deviation leads to an upwards adjustment Coal – estimated coefficients coefficient (percentage) 20 15 10 5 0 -5 -10 -15 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Gas – estimated coefficients coefficient (percentage) 12 10 8 6 4 2 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Smoothing the coefficients for coal Coefficients for coal 20 Coefficients for gas 15 12 10 coefficient (%) coefficient (%) 10 5 0 8 6 4 2 -5 0 Jan -10 -15 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Ja n95 Ju l-9 5 Ja n96 Ju l-9 6 Ja n97 Ju l-9 7 Ja n98 Ju l-9 8 Ja n99 Ju l-9 9 Ja n00 Ju l-0 0 Ja n01 Ju l-0 1 Ja n02 Ju l-0 2 Ja n03 Ju l-0 3 Ja n04 Ju l-0 4 Ja n05 Ju l-0 5 Ja n06 Ju l-0 6 Ja n07 Ju l-0 7 Ja n08 Ju l-0 8 thousands of tonnes Comparing seasonal adjustments Coal consumption, seasonally adjusted 8000 proposed new factors 7500 7000 current method of temperature correction 6500 6000 5500 5000 4500 4000 3500 3000 Heating degree days • The difference between the maximum temperature in a day and some target temperature • If the temperature in one day is above the target then the degree day measure is zero for that day • The choice of target temperature is important Smoothing the coefficients, heating degree days model (Eurostat measure) Coefficients for gas 0.20 0.15 correction factor, per unit deviation from the average degree day correction factor, per unit deviation from the average degree day Coefficients for coal 0.10 0.05 0.00 -0.05 -0.10 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 Jan Feb Mar Apr May Jun -0.15 -0.20 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jul Aug Sep Oct Nov Dec 8 08 l-0 Ju 7 07 l-0 n- Ja Ju 6 06 l-0 n- Ja Ju 5 05 4 l-0 n- Ja Ju n- l-0 04 8000 Ja Ju 3 03 l-0 n- Ja Ju 2 02 1 l-0 n- Ja Ju n- Ja 01 0 l-0 Ju n- Ja 00 9 l-0 Ju n- Ja 99 8 l-9 Ju n- Ja 98 l-9 Ju 7 97 l-9 n- Ja Ju 6 96 l-9 n- Ja Ju 5 95 l-9 n- Ja Ju n- Ja thousands of tonnes Effect on coal seasonal adjustment Coal consumption, seasonally adjusted using Eurostat degree days 7500 7000 current method of temperature correction 6500 6000 5500 5000 4500 4000 3500 3000 The difference temperature correction can make! Million tonnes of oil equivalent Primary energy Temperature consumption Unadjusted 2009 211.1 adjusted 212.6 2010 217.3 211.3 Annual change +2.9% -0.6%