Econometric methods and applications in modelling non-stationary climate data
<p>Understanding of climate change and policy responses thereto rely on accurate measurements as well as models of both socio-economic and physical processes. However, data to assess impacts and establish historical climate records are non-stationary: distributions shift over time due to shock...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2015
|
| Subjects: |
| _version_ | 1826305077594292224 |
|---|---|
| author | Pretis, F |
| author2 | Hendry, D |
| author_facet | Hendry, D Pretis, F |
| author_sort | Pretis, F |
| collection | OXFORD |
| description | <p>Understanding of climate change and policy responses thereto rely on accurate measurements as well as models of both socio-economic and physical processes. However, data to assess impacts and establish historical climate records are non-stationary: distributions shift over time due to shocks, measurement changes, and stochastic trends - all of which invalidate standard statistical inference. This thesis establishes econometric methods to model non-stationary climate data consistent with known physical laws, enabling joint estimation and testing, develops techniques for the automatic detection of structural breaks, and evaluates socio-economic scenarios used in long-run climate projections.</p> <p>Econometric cointegration analysis can be used to overcome inferential difficulties stemming from stochastic trends in time series, however, cointegration has been criticised in climate research for lacking a physical justification for its use. I show that physical two-component energy balance models of global mean climate can be mapped to a cointegrated system, making them directly testable, and thereby provide a physical justification for econometric methods in climate research.</p> <p>Automatic model selection with more variables than observations is introduced in modelling concentrations of atmospheric CO<sub>2</sub>, while controlling for outliers and breaks at any point in the sample using impulse indicator saturation. Without imposing the inclusion of variables <em>a-priori</em>, model selection results find that vegetation, temperature and other natural factors alone cannot explain the trend or the variation in CO<sub>2</sub> growth. Industrial production components, driven by business cycles and economic shocks, are highly significant contributors.</p> <p>Generalizing the principle of indicator saturation, I present a methodology to detect structural breaks at any point in a time series using designed functions. Selecting over these break functions at every point in time using a general-to-specific algorithm, yields unbiased estimates of the break date and magnitude. Analytical derivations for the split-sample approach are provided under the null of no breaks and the alternative of one or more breaks. The methodology is demonstrated by detecting volcanic eruptions in a time series of Northern Hemisphere mean temperature derived from a coupled climate simulation spanning close to 1200 years.</p> <p>All climate models require socio-economic projections to make statements about future climate change. The large span of projected temperature changes then originates predominantly from the wide range of scenarios, rather than uncertainty in climate models themselves. For the first time, observations over two decades are available against which the first sets of socio-economic scenarios used in the Intergovernmental Panel on Climate Change reports can be assessed. The results show that the growth rate in fossil fuel CO<sub>2</sub> emission intensity (fossil fuel CO2 emissions per GDP) over the 2000s exceeds all main scenario values, with the discrepancy being driven by underprediction of high growth rates in Asia. This underestimation of emission intensity raises concerns about achieving a world of economic prosperity in an environmentally sustainable fashion.</p> |
| first_indexed | 2024-03-07T06:27:23Z |
| format | Thesis |
| id | oxford-uuid:f4c9122b-5270-4b55-a292-2cdf10ad7f2a |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T06:27:23Z |
| publishDate | 2015 |
| record_format | dspace |
| spelling | oxford-uuid:f4c9122b-5270-4b55-a292-2cdf10ad7f2a2022-03-27T12:22:23ZEconometric methods and applications in modelling non-stationary climate dataThesishttp://purl.org/coar/resource_type/c_db06uuid:f4c9122b-5270-4b55-a292-2cdf10ad7f2aEconometricsEconomicsEnglishOxford University Research Archive - Valet2015Pretis, FHendry, D<p>Understanding of climate change and policy responses thereto rely on accurate measurements as well as models of both socio-economic and physical processes. However, data to assess impacts and establish historical climate records are non-stationary: distributions shift over time due to shocks, measurement changes, and stochastic trends - all of which invalidate standard statistical inference. This thesis establishes econometric methods to model non-stationary climate data consistent with known physical laws, enabling joint estimation and testing, develops techniques for the automatic detection of structural breaks, and evaluates socio-economic scenarios used in long-run climate projections.</p> <p>Econometric cointegration analysis can be used to overcome inferential difficulties stemming from stochastic trends in time series, however, cointegration has been criticised in climate research for lacking a physical justification for its use. I show that physical two-component energy balance models of global mean climate can be mapped to a cointegrated system, making them directly testable, and thereby provide a physical justification for econometric methods in climate research.</p> <p>Automatic model selection with more variables than observations is introduced in modelling concentrations of atmospheric CO<sub>2</sub>, while controlling for outliers and breaks at any point in the sample using impulse indicator saturation. Without imposing the inclusion of variables <em>a-priori</em>, model selection results find that vegetation, temperature and other natural factors alone cannot explain the trend or the variation in CO<sub>2</sub> growth. Industrial production components, driven by business cycles and economic shocks, are highly significant contributors.</p> <p>Generalizing the principle of indicator saturation, I present a methodology to detect structural breaks at any point in a time series using designed functions. Selecting over these break functions at every point in time using a general-to-specific algorithm, yields unbiased estimates of the break date and magnitude. Analytical derivations for the split-sample approach are provided under the null of no breaks and the alternative of one or more breaks. The methodology is demonstrated by detecting volcanic eruptions in a time series of Northern Hemisphere mean temperature derived from a coupled climate simulation spanning close to 1200 years.</p> <p>All climate models require socio-economic projections to make statements about future climate change. The large span of projected temperature changes then originates predominantly from the wide range of scenarios, rather than uncertainty in climate models themselves. For the first time, observations over two decades are available against which the first sets of socio-economic scenarios used in the Intergovernmental Panel on Climate Change reports can be assessed. The results show that the growth rate in fossil fuel CO<sub>2</sub> emission intensity (fossil fuel CO2 emissions per GDP) over the 2000s exceeds all main scenario values, with the discrepancy being driven by underprediction of high growth rates in Asia. This underestimation of emission intensity raises concerns about achieving a world of economic prosperity in an environmentally sustainable fashion.</p> |
| spellingShingle | Econometrics Economics Pretis, F Econometric methods and applications in modelling non-stationary climate data |
| title | Econometric methods and applications in modelling non-stationary climate data |
| title_full | Econometric methods and applications in modelling non-stationary climate data |
| title_fullStr | Econometric methods and applications in modelling non-stationary climate data |
| title_full_unstemmed | Econometric methods and applications in modelling non-stationary climate data |
| title_short | Econometric methods and applications in modelling non-stationary climate data |
| title_sort | econometric methods and applications in modelling non stationary climate data |
| topic | Econometrics Economics |
| work_keys_str_mv | AT pretisf econometricmethodsandapplicationsinmodellingnonstationaryclimatedata |