diff --git a/.gitignore b/.gitignore index 63d5a56..782274f 100644 --- a/.gitignore +++ b/.gitignore @@ -139,3 +139,4 @@ dmypy.json /project/input/_old/ /project/input/abattement_curve/data_preparation.xlsx /documentation/_old/ +/project/input/phebus/data_preparation_phebus.xlsx diff --git a/.idea/workspace.xml b/.idea/workspace.xml index 2784691..c0fd535 100644 --- a/.idea/workspace.xml +++ b/.idea/workspace.xml @@ -1,47 +1,99 @@ - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + - + + + + + + + + + + + + + + + + + + + + + + + + + + + - + + + + + + + @@ -93,7 +145,7 @@ - + @@ -103,21 +155,21 @@ - - + + + + - - - + + @@ -426,7 +479,6 @@ - @@ -604,14 +656,9 @@ - - - - 1625695396212 - 1625750755080 @@ -949,7 +996,14 @@ - @@ -1023,10 +1077,10 @@ - + - + @@ -1035,9 +1089,9 @@ - - - + + + diff --git a/docs/LICENSE.html b/docs/LICENSE.html index c55a886..98bef88 100644 --- a/docs/LICENSE.html +++ b/docs/LICENSE.html @@ -94,9 +94,12 @@
  • License
  • Input Res-IRF version 3.0
  • Technical documentation
    • -
  • Influence of input variables
    • +
  • Simulation and sensitivity analysis
    • +
  • Setting up public policies
    • +
  • Policies assessment
  • Module
    • -
  • Additional information
    • +
  • Modification from Res-IRF 3.0
    • +
  • Installation help
  • Contributing
  • Future developments
    • diff --git a/docs/README.html b/docs/README.html index dea9bb0..b724ade 100644 --- a/docs/README.html +++ b/docs/README.html @@ -103,9 +103,12 @@
  • License
  • Input Res-IRF version 3.0
  • Technical documentation
    • -
  • Influence of input variables
    • +
  • Simulation and sensitivity analysis
    • +
  • Setting up public policies
    • +
  • Policies assessment
  • Module
    • -
  • Additional information
    • +
  • Modification from Res-IRF 3.0
    • +
  • Installation help
  • Contributing
  • Future developments
    • diff --git a/docs/_images/agents_comparison_rate.png b/docs/_images/agents_comparison_rate.png new file mode 100644 index 0000000..8197359 Binary files /dev/null and b/docs/_images/agents_comparison_rate.png differ diff --git a/docs/_images/agents_different_rate.png b/docs/_images/agents_different_rate.png new file mode 100644 index 0000000..825bb1d Binary files /dev/null and b/docs/_images/agents_different_rate.png differ diff --git a/docs/_images/agents_same_rate.png b/docs/_images/agents_same_rate.png new file mode 100644 index 0000000..7f77fe3 Binary files /dev/null and b/docs/_images/agents_same_rate.png differ diff --git a/docs/_images/consumption_actual.png b/docs/_images/consumption_actual.png index 8cb2843..06d5fc0 100644 Binary files a/docs/_images/consumption_actual.png and b/docs/_images/consumption_actual.png differ diff --git a/docs/_images/consumption_actual1.png b/docs/_images/consumption_actual1.png index 7e22bb6..8cb2843 100644 Binary files a/docs/_images/consumption_actual1.png and b/docs/_images/consumption_actual1.png differ diff --git a/docs/_images/consumption_actual2.png b/docs/_images/consumption_actual2.png index 90d3e9d..7e22bb6 100644 Binary files a/docs/_images/consumption_actual2.png and b/docs/_images/consumption_actual2.png differ diff --git a/docs/_images/consumption_actual3.png b/docs/_images/consumption_actual3.png new file mode 100644 index 0000000..90d3e9d Binary files /dev/null and b/docs/_images/consumption_actual3.png differ diff --git a/docs/_images/flow_renovation.png b/docs/_images/flow_renovation.png index eae9243..0d4a0ca 100644 Binary files a/docs/_images/flow_renovation.png and b/docs/_images/flow_renovation.png differ diff --git a/docs/_images/flow_renovation1.png b/docs/_images/flow_renovation1.png index 0b5a02d..eae9243 100644 Binary files a/docs/_images/flow_renovation1.png and b/docs/_images/flow_renovation1.png differ diff --git a/docs/_images/flow_renovation2.png b/docs/_images/flow_renovation2.png index fa174a6..0b5a02d 100644 Binary files a/docs/_images/flow_renovation2.png and b/docs/_images/flow_renovation2.png differ diff --git a/docs/_images/flow_renovation3.png b/docs/_images/flow_renovation3.png new file mode 100644 index 0000000..fa174a6 Binary files /dev/null and b/docs/_images/flow_renovation3.png differ diff --git a/docs/_images/function_agents.png b/docs/_images/function_agents.png new file mode 100644 index 0000000..e125265 Binary files /dev/null and b/docs/_images/function_agents.png differ diff --git a/docs/_images/renovation_fitting_function.png b/docs/_images/renovation_fitting_function.png new file mode 100644 index 0000000..80f56ba Binary files /dev/null and b/docs/_images/renovation_fitting_function.png differ diff --git a/docs/_images/renovation_rate_decision_maker.png b/docs/_images/renovation_rate_decision_maker.png new file mode 100644 index 0000000..ae27595 Binary files /dev/null and b/docs/_images/renovation_rate_decision_maker.png differ diff --git a/docs/_images/renovation_rate_performance.png b/docs/_images/renovation_rate_performance.png new file mode 100644 index 0000000..bf974a3 Binary files /dev/null and b/docs/_images/renovation_rate_performance.png differ diff --git a/docs/_sources/changelog.md.txt b/docs/_sources/changelog.md.txt new file mode 100644 index 0000000..357a246 --- /dev/null +++ b/docs/_sources/changelog.md.txt @@ -0,0 +1,203 @@ + +# Modification from Res-IRF 3.0 + +Modifications are listed in the table and the major changes are developed if necessary in a separate section. + +```{eval-rst} +.. csv-table:: Code modification + :name: code_modificaiton + :file: table/changelog_resirf.csv + :header-rows: 1 +``` + +## Renovation rate function + +The renovation rate $τ_i$ of dwellings labelled i is then calculated as a logistic function of the NPV: + +$$τ_i=\frac{τ_{max}}{(1+(τ_{max}/τ_{min} -1) e^{-ρ(NPV_i- NPV_{min})})}$$ + +- $τ_{max}$ and NPV_{min} are constant based on own assumption, +- $NPV_i$ is a result from Res-IRF, +- $ρ$ is calibrated to replicate the observed renovation rates. + +Previous work {cite:ps}`brangerGlobalSensitivityAnalysis2015` showed that the model was very sensitive to this parameter. + +Res-IRF 3.0, based on the Phébus building stock, replicated the renovation rates observed in the French housing stock +according to the Energy Performance Certificate (EPC) and the Decision Maker (DM). + +Renovation rates by pair (ECD, decision maker) were determined using the following data: +- 3% of all dwellings in the stock are renovated in the original year (source ADEME); +- Share of renovations performed for each EPC {numref}`renovation_share`: $P(q | R)$ +- Renovation rate by decision maker[^maker] {numref}`renovation_rate` : $P(R | d)$ + +[^maker]: Parameters imported in the Scilab version of Res-IRF were different from those indicated in some articles. We +present here the correct values. + +```{eval-rst} +.. csv-table:: Renovation share by energy performance certificate. Source: PUCA (2015) + :name: renovation_share + :file: table/input_2012/renovation_share_ep_2012.csv + :header-rows: 1 + :stub-columns: 1 +``` + +```{eval-rst} +.. csv-table:: Renovation rate by decision make. Source : OPEN (2016) and USH (2017) + :name: renovation_rate + :file: table/input_2012/renovation_rate_dm_2012.csv + :header-rows: 1 + :stub-columns: 1 +``` + +### Correction of renovation rate objective formula + +The usual conditional probability formulas allow us to calculate rate_obj (renovation rate as a function of decision +maker and EPC) as follows: + +$$\text{rate_obj} = P(R | d \cap q) = \frac{P(R \cap d \cap q)}{P(d \cap q)} = \frac{P(R) P(d \cap q | R)}{P(d \cap q)}$$ + +Assuming independence between: + +$$P(d \cap q | R) = P(q|R) P(d | q \cap R) = P(q | R) P(d|R)$$ + + +$$\text{rate_obj} = P(R | d \cap q) = \frac{P(d) P(q) P(R|d) P(R|q)}{P(R) P(d \cap q)}$$ + +- $P(R)$ : aggregate renovation rate of the stock of 3%. +- $P(R|d)$ in {numref}`renovation_rate` +- $P(R|q) = P(R \cap q) / P(q) = P(R) * P(q|R) / P(q)$ and with $P(q|R)$ in {numref}`renovation_share` +- $P(q)$, $P(d)$, $P(d \cap q)$ calculated by the building stock structure. + + +Previously Res-IRF 3.0 calculated : + +$$\text{rate_obj} = \frac{P(R|d) P(R|q)}{P(R)}$$ + +**We replace this formula by the formula defined above.** We assess the marginal impact of this modification on Res-IRF +3.0. We observe a change in absolute value but not in trend. + + +```{figure} img/changelog/renovation_rate_formula/flow_renovation.png +:name: renovation_rate_formula_flow_renovation + +Impact renovation rate formula on flow_renovation +``` + +```{figure} img/changelog/renovation_rate_formula/renovation_rate_decision_maker.png +:name: renovation_rate_formula_renovation_rate_decision_maker + +Impact renovation rate formula on decision-maker renovation rate +``` + +```{figure} img/changelog/renovation_rate_formula/renovation_rate_performance.png +:name: renovation_rate_formula_renovation_rate_performance + +Impact renovation rate formula on EPC renovation rate +``` + +```{figure} img/changelog/renovation_rate_formula/consumption_actual.png +:name: renovation_rate_formula_consumption_actual + +Impact renovation rate formula on consumption actual +``` + +### Calibration segmentation + +Parameter ρ is calibrated, for each type of decision-maker and each initial label, so that the NPVs calculated with the +subsidies in effect in 2012 (version 3.0) reproduce the observed renovation rates. In general, we don't have enough +disaggregated data. +**How can we calibrate an individual decision function per agent with only aggregated data?** + +A simple solution would be to say that each agent must replicate the observed renovation rates of his group. In this +case, we have as many decision functions as there are agents. This solution is not satisfactory because it erases all +the specificities of the agents (discount rate, investment horizon, etc...) and is equivalent to creating a model with a +representative agent. Also, a robust decision function must be unique for all agents. + +```{figure} img/changelog/renovation_rate_function/function_agents.png +:name: function_agents + +Individual decision functions by agents. +``` + +#### Agents with the same observed renovation rate should get the same renovation rate function + +Let say two agents got 2 NPV of 100 €/m2 (Agent 1) and 200 €/m2 (Agent 2) and their observed renovation rate is 3%. This +occurs when, due to lack of data, we assign two different agents the same renovation rate because they share common +attributes. +- For example, Agent 1 and Agent 2 are both Homeowners in a Single-family dwelling, but one is living +in high-efficient dwelling when the other is living in a low-efficient dwelling. Agent 1 should get lower incentive to +invest than Agent 2. +- Another example, will be between 2 agents of 2 different owner income class. Agent 1 owner could be +D1 with expensive investment loan when Agent 2 could become to D10. + +```{figure} img/changelog/renovation_rate_function/agents_same_rate.png +:name: agents_same_rate + +Individual decision functions by agents of the same group. +``` + +These situations push us to create one renovation rate function for all agents sharing the same observed renovation rate, +this means the same attributes. The objective is to determine the parameters in such a way that the average weighted by +the number of agents reproduces the observed renovation rate. + +The ideal solution would be to determine the function parameter by solving the following equation: + +$$\sum_{k=1}^n w(k) f(ρ, NPV_k) = \text{rate_obj}$$ + +However, this solution is not analytically solvable. + +We are thinking of two other ways to calibrate the renovation function for all agents sharing the same attributes: +- Calculate function parameter for each individual agent and then calculate an average parameter. +- Calculate function parameter for a fictive individual agent that got an average utility function. + +```{figure} img/changelog/renovation_rate_function/agents_comparison_rate.png +:name: agents_comparison_rate + +Comparison of decision functions with the same renovation rate objective. +``` + +- There isn't critical differences between the two methods. +- There is no indication that the weighted average of the parameters allows us to recover the average value of the + observed renovation rate. This is an approximation. +- Although without any theoretical support, we think it is simpler to imagine the average observed renovation rate + representing the decision of a representative agent. + +We will therefore calibrate the decision function by the utility of a representative agent. + +#### All agents should get the same renovation rate function + +Should agents with different observed renovation rate get different renovation rate function? + +```{figure} img/changelog/renovation_rate_function/agents_different_rate.png +:name: agents_different_rate + +Should agents with different observed renovation rate get different renovation rate function? +``` + +We believe that it is more robust to determine a single decision function for all agents. We thus define the decision +function as the best logistic function passing through the pairs (Utility, Investment rate) or (NPV, Renovation rate) +for our specific case. + +Let say two agents with the following attributes: +Agent 1: NPV of 100 €/m2, and observed renovation rate of 2% +Agent 2: NPV of 200 €/m2, and observed renovation rate of 3%. + +```{figure} img/changelog/renovation_rate_function/renovation_fitting_function.png +:name: renovation_fitting_function + +Fitting renovation rate objective data to utility data calculated. +``` + +#### Conclusion + +We try to replicate the observed investment decision from a utility function per agent calculated by a model. However, +we do not have sufficiently disaggregated data to perform this calibration directly. + +**How can we calibrate an individual decision function per agent with only aggregated data?** + +1. Define a representative agent by segmentation of the observed data, +2. Calculate the utility function for the representative agent, +3. Fitting observed renovation rate data to the utility calculated. + +Even without disaggregated data, each agent will have a different decision when calibrating. The average of the agents +per group of the same attributes will reflect the observed decision rates. \ No newline at end of file diff --git a/docs/_sources/help_installation.md.txt b/docs/_sources/help_installation.md.txt index a7d86ac..b9cb7ca 100644 --- a/docs/_sources/help_installation.md.txt +++ b/docs/_sources/help_installation.md.txt @@ -1,4 +1,4 @@ -# Additional information +# Installation help ## Conda environment ### Creating an environment from an environment.yml file diff --git a/docs/_sources/index.rst.txt b/docs/_sources/index.rst.txt index 2afbac9..25753be 100644 --- a/docs/_sources/index.rst.txt +++ b/docs/_sources/index.rst.txt @@ -15,7 +15,10 @@ Welcome to Res-IRF's documentation! input_2012 technical_documentation simulation_2012 + policies_parameters + policies_assessment modules + changelog help_installation contributing future_development diff --git a/docs/_sources/input_2012.md.txt b/docs/_sources/input_2012.md.txt index 6e7f0f5..0b788d3 100644 --- a/docs/_sources/input_2012.md.txt +++ b/docs/_sources/input_2012.md.txt @@ -38,7 +38,7 @@ reference for most projections of energy consumption in France. ```{eval-rst} .. csv-table:: Overview of the database :name: databases_overview - :file: table/databases_overview.csv + :file: table/input_2012/databases_overview_2012.csv :header-rows: 1 :stub-columns: 1 @@ -48,7 +48,7 @@ reference for most projections of energy consumption in France. ```{eval-rst} .. csv-table:: Overview of the database content :name: databases_overview_content - :file: table/databases_overview_content.csv + :file: table/input_2012/databases_overview_content_2012.csv :header-rows: 1 :stub-columns: 1 @@ -94,7 +94,7 @@ multi-family dwellings) and types of investors (owner-occupied, landlord, social ```{eval-rst} .. csv-table:: Joint distribution of building and investor characteristics in Res-IRF 3.0 :name: decision_maker_distribution - :file: table/decision_maker_distribution.csv + :file: table/input_2012/decision_maker_distribution_phebus.csv :header-rows: 1 :stub-columns: 1 @@ -152,7 +152,7 @@ Distribution of income categories within EPC bands. Source: Phébus ```{eval-rst} .. csv-table:: Income categories used in Res-IRF 3.0 - :file: table/income_categories.csv + :file: table/input_2012/income_categories_phebus.csv :name: income_categories :header-rows: 1 :stub-columns: 1 @@ -199,7 +199,7 @@ in the model. Model inputs fall into three categories {cite:ps}`brangerGlobalSen ```{eval-rst} .. csv-table:: Complete list of inputs :name: input_list_2012 - :file: table/input_list_2012.csv + :file: table/input_2012/input_list_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -229,7 +229,7 @@ Market shares used to calibrate intangible costs for construction. ```{eval-rst} .. csv-table:: Market shares of construction in 2012. :name: market_share_construction_2012 - :file: table/ms_construction_ini_2012.csv + :file: table/input_2012/ms_construction_ini_2012.csv :header-rows: 2 :stub-columns: 2 ``` @@ -242,7 +242,7 @@ shares reported here {numref}`market_share_2012` ```{eval-rst} .. csv-table:: Market shares of energy efficiency upgrades in 2012. Source: PUCA (2015) :name: market_share_2012 - :file: table/ms_renovation_ini_2012.csv + :file: table/input_2012/ms_renovation_ini_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -260,7 +260,7 @@ their aggregation represents 3% (686,757 units) of the housing stock of the init ```{eval-rst} .. csv-table:: Renovation share by energy performance label. Source: PUCA (2015) :name: renovation_share_ep_2012 - :file: table/renovation_share_ep_2012.csv + :file: table/input_2012/renovation_share_ep_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -268,7 +268,7 @@ their aggregation represents 3% (686,757 units) of the housing stock of the init ```{eval-rst} .. csv-table:: Renovation rate by type of dwelling. Source : OPEN (2016) and USH (2017) :name: renovation_rate_dm_2012 - :file: table/renovation_rate_dm_2012.csv + :file: table/input_2012/renovation_rate_dm_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -282,7 +282,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Calibration of total final actual energy consumption :name: calibration_energy_2012 - :file: table/ceren_energy_consumption_2012.csv + :file: table/input_2012/ceren_energy_consumption_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -292,7 +292,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Initial Floor area construction (m2/dwelling) :name: area_construction_2012 - :file: table/area_construction_2012.csv + :file: table/input_2012/area_construction_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -300,7 +300,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Floor area construction elasticity :name: area_construction_elasticity_2012 - :file: table/area_construction_elasticity_2012.csv + :file: table/input_2012/area_construction_elasticity_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -308,7 +308,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Maximum Floor area construction (m2/dwelling) :name: area_construction_max_2012 - :file: table/area_construction_max_2012.csv + :file: table/input_2012/area_construction_max_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -317,7 +317,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Rotation and Mutation rate (%/year) :name: rotation_mutation_rate_2012 - :file: table/rotation_mutation_rate_2012.csv + :file: table/input_2012/rotation_mutation_rate_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -327,7 +327,7 @@ conversion coefficients applied to the Phebus Building Stock are listed in {numr ```{eval-rst} .. csv-table:: Renovation costs used in Res-IRF 3.0 (€/m2). Source: Expert opinion :name: cost_renovation_2012 - :file: table/cost_renovation_2012.csv + :file: table/input_2012/cost_renovation_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -340,7 +340,7 @@ model. ```{eval-rst} .. csv-table:: Switching-fuel costs used in Res-IRF 3.0 (€/m2). Source: Expert opinion :name: switching_fuel_cost - :file: table/cost_switch_fuel_2012.csv + :file: table/input_2012/cost_switch_fuel_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -348,7 +348,7 @@ model. ```{eval-rst} .. csv-table:: Construction costs (€/m2) :name: cost_construction_2012 - :file: table/cost_construction_2012.csv + :file: table/input_2012/cost_construction_2012.csv :header-rows: 2 :stub-columns: 1 ``` @@ -358,7 +358,7 @@ model. ```{eval-rst} .. csv-table:: Initial Floor area (m2/dwelling) :name: area_existing_2012 - :file: table/area_existing_2012.csv + :file: table/input_2012/area_existing_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -366,7 +366,7 @@ model. ```{eval-rst} .. csv-table:: Income (€/year) :name: income_2012 - :file: table/income_2012.csv + :file: table/input_2012/income_2012.csv :header-rows: 1 ``` @@ -375,7 +375,7 @@ model. ```{eval-rst} .. csv-table:: Discount rates (%/year). Source: Expert opinion :name: discount_rate_existing_2012 - :file: table/discount_rate_existing_2012.csv + :file: table/input_2012/discount_rate_existing_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -383,7 +383,7 @@ model. ```{eval-rst} .. csv-table:: Investment horizon (years). Source: Expert opinion :name: investment_horizon_2012 - :file: table/investment_horizon_2012.csv + :file: table/input_2012/investment_horizon_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -395,7 +395,7 @@ discount rate of 7% and a time horizon of 35 years. ```{eval-rst} .. csv-table:: Discount rates construction (%/year). Source: Expert opinion :name: discount_rate_construction_2012 - :file: table/discount_rate_construction_2012.csv + :file: table/input_2012/discount_rate_construction_2012.csv :header-rows: 1 :stub-columns: 1 ``` diff --git a/docs/_sources/policies_assessment.md.txt b/docs/_sources/policies_assessment.md.txt new file mode 100644 index 0000000..5ce5090 --- /dev/null +++ b/docs/_sources/policies_assessment.md.txt @@ -0,0 +1,139 @@ + +# Policies assessment + +## General principles + +The simulations are based on the following principles: + +1. All instruments apply from 2012 (except the carbon tax which applies from 2014) and are maintained until 2050; they + are therefore taken into account in the calibration of the model; +2. The different instruments are fully cumulative with each other; +3. The instruments work at full capacity; in particular, a household that carries out a renovation +4. The instruments work at full capacity; in particular, a household that carries out a renovation receives all the aid + for which it is eligible; +5. The instruments apply equally to private and social housing[^housing]; +6. Incentives are perfectly passed on to households, without being captured by energy efficiency vendors in the form of + an increase in the base price; this mechanism is based on an assumption of perfect competition in the energy + renovation sector (Nauleau et al., 2015). These assumptions are relatively optimistic. As detailed in the paragraphs + below, however, policy-specific parameterization elements moderate this optimism. More generally, the study aims at + least as much to qualitatively illustrate the mechanisms by which the different instruments operate as to provide a + precise quantitative assessment of their impact. Finally, it is important to remember that integrated energy-economy + modeling aims to understand mechanisms and quantify orders of magnitude rather than to provide precise results - as + summarized by the formula modeling for insights, not numbers advocated by the Energy Modeling Forum (Huntington et + al., 1982). In particular, the inaccuracies of the modeling, i.e. the discrepancies that may exist between + simulations and observations, can be instructive in the sense that they highlight gaps in the knowledge of the + processes, thus making it possible to orientate the research priorities. + +[^housing]: This hypothesis is particularly optimistic, since in practice the social housing stock is only concerned by +the CEE and the VATr. However, it has little influence, as shown by the results presented below through the low +sensitivity of the renovation of social housing to the level of aid. + +## Evaluation Indicators + +Version 3.0 of Res-IRF allows for a multi-criteria evaluation of policies. The effectiveness of an intervention - +instrument or mix of instruments - is assessed as its ability to achieve the objectives assigned to it - in this case, +the five objectives defined bellow. To this indicator, which was the core of the evaluation conducted in 2011 for the +CGDD, is now added the efficiency of an intervention, assessed as its capacity to achieve a certain objective at the +lowest cost. This notion only applies to incentive instruments and therefore not to thermal regulations. Two efficiency +indicators are calculated: the **cost effectiveness** and the **leverage effect**. Finally, the differentiation of +households by income category allows us to evaluate the distributional effects. + +### Effectiveness + +Evaluating the additional effect of an intervention requires the formulation of a counterfactual scenario, without the +intervention under consideration. One of the advantages of the Res-IRF modeling architecture is the simplicity with +which an intervention can be added or removed, facilitating the construction of a counterfactual scenario. An important +difficulty, however, is that the multiplicity of instruments represented opens the door to a multiplicity of +counterfactual scenarios. Specifically, within a bundle of n instruments (at a given parameterization), each instrument +can be evaluated alone, or in interaction with 1 to n - 1 instruments; with n = 6, it is thus possible to simulate 64 +scenarios. In the exercise presented here, four scenarios receive particular attention: + +- Trend scenario with all policies (AP): includes all instruments in their default variant. This is the scenario on + which the model is calibrated at the starting year (2012). +- Counterfactual scenario without any policies (ZP for "zero policies"): all instruments are removed [^removed], without + recalibrating the model. The initial energy consumptions are therefore not the initial energy consumptions are + therefore not reproduced exactly. +- Trend scenario + (AP+): includes all instruments in their "+" variant [^variant]. +- AP scenario without landlord-tenant dilemma (DPL): the behavior of landlords is modelled on that of owner-occupiers (" + full green value" scenario, {numref}`investment_horizon_2012`). This variant aims to quantify the energy savings that + would be generated by the different instruments if they were perfectly targeted at the rental stock [^stock]. + +[^removed]: Except for the reduced VAT rate, which is replaced by the standard 10% VAT rate. +[^variant]: This scenario leads, at the beginning of the period, to a few situations where households occupying housing +classified as G or F receive a total subsidy of more than 100%. +[^stock]: This implies an increased promotion of the various aids to landlords and a specific accompaniment of the work. + +Among the 64 possible instrument combinations, the additional effect of each policy can be evaluated in 32 different +ways, depending on the counterfactual situation considered. Two counterfactual situations, obtained by two different +methods, are of particular interest because they make it possible to limit the impact of each instrument: + +- AP-1 method: the AP scenario is compared to an alternative scenario without the instrument considered. The difference + between the two scenarios gives the impact of the instrument in interaction with all other instruments. +- ZP+1" method: the ZP scenario is compared to an alternative scenario with the instrument considered. The difference + between the two scenarios gives the pure impact of the instrument. + +For each instrument, the comparison of the impacts obtained by the two methods makes it possible to evaluate the +importance of its interactions with all the other instruments. Nevertheless, for a better readability of the results, in +the rest of the documentation the estimates in AP-1 are preferred, which correspond to the counterfactual situation that +is closest to reality, in the sense that it is based on a minimum of assumptions. + +### Efficiency + +The efficiency of the incentive instruments is evaluated here through the indicators of cost-effectiveness and leverage. +To estimate the marginal effect of the instrument at a year 𝑡, scenarios with and without the instrument are compared +at year 𝑡. Since this method is computationally intensive, we limit ourselves to the cut-off points 2015, 2025 and +2035. + +#### Cost-Effectiveness (CE) + +The cost-effectiveness indicator relates the costs of the incentive to its effectiveness measured in terms of energy +savings. The cost considered corresponds to the tax expenditure in the case of subsidies and the tax revenue in the case +of the tax (i.e., negative cost). The indicator is calculated here in conventional and real energy. The "conventional +energy" metric makes it possible to avoid the heterogeneous behavioral effects between households, which vary greatly +from one year to the next with short-term fluctuations in energy prices. Compared to the "real energy" metric, it leads +to an overestimation of the effectiveness of subsidies by ignoring the rebound effect they generate, and to an +underestimation of the effectiveness of energy taxes, which on the contrary induce a sobriety effect. The two metrics +should produce similar results for WHO, a hybrid instrument. To calculate the cost-effectiveness indicator $CE$, the +energy savings $\Delta E_{t}$ between the scenarios with the incentive present or absent at year $t$ (but in both cases +present until year $t-1$) are compared, applying a discount factor DF: + +$$CE_{t} = \frac{\text{Incitation}_{t}}{\Delta E_{t} DF}$$ + +Using a discount rate of 4% and a lifetime of 26 years, which corresponds to the average of the operations carried out +in the framework of the WHO on the residential building perimeter, the factor $DT$ is taken to be equal to 16.6. + +#### Leverage Effect (LE) + +The leverage effect LE relates the effectiveness of the instrument, measured in terms of capital expenditure, to the +cost of the incentive. A leverage effect equal to 1 implies that one euro of public money (grant expenditure or tax +revenue) induces an additional investment of one euro. The leverage effect synthesizes several effects, and its value +will depend on the relative share of "additional" and "non-additional" participants: + +- The incentive benefits "additional" participants, who would not have invested without it. For these individuals, the + leverage effect is much greater than 1, since the subsidy is literally the trigger for investment. For example, a 10% + grant results in a leverage effect equal to 10 - a €100 investment is triggered by a €10 grant. +- The incentive also benefits (and, in our modeling, fully benefits) "non-additional" ("infra-marginal" in economic + terms) participants, who would have invested without the incentive. While this effect is usually referred to as a " + dead weight loss", it is not a pure loss in our model as participants adjust their investment choices towards more + costly options in response to the incentive. Because non-additional participants have heterogeneous preferences, the + leverage effect may be greater than one for some and close to zero for others. + +The formula used applies to all participants and relates the cost of the incentive to the surplus of investment +$\Delta Inv_{t}$ induced by the policy, measured as the difference between two scenarios with the incentive absent or +present at year $t$, but in both cases present until year $t - 1$: + +$$EL_{t} = \frac{\Delta Inv_{t}}{\text{Incitation}_{t}}$$ + + +### Fuel poverty + +Fuel poverty is measured according to the energy effort rate (EER) indicator that has been a reference in Europe in +recent years (Hills, 2012, p.30), counting the number of households that spend more than 10% of their income on energy +expenses for heating, measured in relation to conventional energy consumption, given in France by the 3CL method of the +DPE. This indicator covers 2.7 million households in 2012. For comparison, the Observatoire de la précarité +énergétique (ONPE, 2016) counts 2.8 million households according to a similar indicator but applied to actual energy +consumption and restricted to the first three deciles of the income distribution. In the present exercise, we favor the +conventional consumption indicator, on the grounds that the actual consumption indicator would not account for certain +restrictions in heating behavior that nevertheless constitute a form of fuel poverty. In complement, one can also +examine how the intensity of heating system use, which reflects indoor comfort, varies with income category. This +indicator can be interpreted as an indirect measure of the ONPE's subjective cold indicator (FR). diff --git a/docs/_sources/policies_parameters.md.txt b/docs/_sources/policies_parameters.md.txt new file mode 100644 index 0000000..cade984 --- /dev/null +++ b/docs/_sources/policies_parameters.md.txt @@ -0,0 +1,162 @@ + +# Setting up public policies + +The parameterization elements detailed below are summarized {numref}. + +```{eval-rst} +.. csv-table:: Summary of key policy parameters. + :name: public_policies_summary + :file: table/policies/policies_summary.csv + :header-rows: 1 + :stub-columns: 1 +``` + +## Carbon tax (CAT) + +The carbon tax is applied as of 2014 to natural gas and heating oil. Its revenues are not recycled to +households. The instrument is parameterized as follows: +- Its rate (€/tCO2) is the same as in the TECV law: €30.50 in 2017, €39 in 2018, €47.5 in 2019 and 56€ in 2020. From + 2021 onwards, a growth rate of 6%/year will be applied to reach the target value of €100 in 2030. The tax rate then + evolves at a rate of 4%/year, as recommended by the Quinet report (2008). This rate is subject to the same 20% VAT + that applies to energy prices. +- The carbon contents to which the tax applies are 271 gCO2/kWh PCI for heating oil and 206 gCO2/kWh PCI for natural + gas. The latter will decrease at a rate of 1%/year from 2020 onwards in order to take into account the objectives of + renewable gas penetration (leading to a 26% share in 2050). + +The purpose of the carbon tax is to send a price signal to investors to redirect long-term investments. In accordance +with this principle, the future rate of the tax is announced several years in advance by the government. Such a tax +takes full effect if investors form perfect expectations - they take into account the chronicity of the tax in their +profitability calculations. In practice, anticipatory behavior is closer to the myopia that prevails in the Res-IRF +model. Therefore, we define the following scenario variants, which limit the impact of the instrument: +- Scenario TC: myopically anticipated carbon tax +- Scenario TC+: perfectly anticipated carbon tax + +## Zero-interest loan (ZIL) + +The ZIL is represented as of 2012 as a subsidy equal to the interest on a consumer credit[^credit] used for an investment of +an equivalent amount. Among the different subsidies evaluated, the ZIL has the particularity of targeting work that +generates substantial jumps in performance. In the model, the targeting is based on the "minimum overall energy +performance" requirements, interpreted as follows: +- Post-work consumption ceiling of 150 kWh/m²/year if pre-work consumption exWCOds 180 kWh/m²/year. Interpreted as a + minimum final D label for jumps from the initial G to E label. +- Consumption ceiling after work of 80 kWh/m²/year if consumption before work is less than 180 kWh/m²/year. Interpreted + as a minimum final label B for jumps from the initial label D and C. + +[^credit]: An alternative would be to take as a reference the interest rates of a real estate loan, which are generally lower than +those of a consumer loan. However, this hypothesis is only relevant for a minority of situations where the renovation +work is coupled with the purchase of the property. + +By construction, the assumptions made about the terms of the consumer credit define the amount of the subsidy. Two +alternative scenarios are selected: + +- ZIL scenario: consumer credit terms are those given by OPEN (2016), i.e., an interest rate of 3% over 5 years. In + addition, in order to incorporate the heterogeneous financing constraints faced by households, we add to each + household category a loan ceiling corresponding to the average amount borrowed under the ZIL in the data of the + Société de Gestion du Fonds de Garantie à l'Action Sociale (SGFGAS), from 16,800€ for categories C1 to 21,000€ for + categories C5. Finally, a minimum loan threshold of €5,000 is added. +- ZIL+ scenario: the instrument is closer to its theoretical setting, defined by at an interest rate of 4% (as noted by + the Banque de France for consumer loans in 201529 ) over a 10-year life (maximum authorized duration of an ZIL), with + a loan amount capped at €30,000 and no minimum threshold. Converted into an ad valorem subsidy, these two variants + correspond to subsidy rates of 9% for the ZIL and 23% for the ZIL+. + +## Tax credit for energy transition + +The CITE is represented from 2012 onwards as an ad valorem subsidy at a single rate of 17%, which corresponds to the +average rate of assistance reported by OPEN30. The difference between this value and the official rate of The difference +between this value and the official rate of 30% reflects the fact that, in the model, the subsidy rate applies to the +full cost of the energy performance (supply and labor) when the official CITE rate applies to the cost of supply only ( +except for opaque wall insulation). The instrument is modeled as a subsidy received immediately by the beneficiary, +without taking into account the maximum 12-month delay inherent in any tax credit31. In order to take into account the +current discussions on restricting the instrument to the most efficient measures (by cancelling the In order to take +into account the current discussions on restricting the instrument to the most efficient measures (by cancelling the +eligibility of windows, for example), two variants are modeled: +ISCED scenario: non-targeted subsidy CITE+ scenario: targeted subsidy like the ZIL + +## White certificate obligations (WCO) + +The WCO are represented from 2012 onwards as a hybrid instrument coupling an energy efficiency subsidy whose cost is +passed on by the obliged energy suppliers as a tax on energy sales. These two components are modeled as follows: + +- The amount of the subsidy is defined for each operation by an amount of discounted cumulative energy savings {numref}`cumac_label_transition` + multiplied by a WCO price (€/kWh cumulated). The latter is the subject of a scenario described below. The operations + considered are those relating to the envelope and thermal systems of residential buildings [^buildings]. + {numref}`example_who_prices` relates the amounts thus calculated to the investment costs for the year 2012 to illustrate + the ad valorem rates associated with such a subsidy. While the average rate is 5%, the WCO scale is "regressive" in + the sense that the highest subsidy rates are for the least expensive operations; indeed, although the amounts granted + increase with the energy performance achieved, this increase is less than the underlying increase in investment costs. + The amount of the WCO subsidy is capped so that the total subsidy rate, including the CITE, TVAr and ZIL, does + not exceed 100%. +- The tax (in € per kWh sold) is calculated by multiplying the official obligation coefficients (kWh cumac per kWh sold) + by the price of WCO (€/kWh cumac). This tax is restricted to sales of domestic fuel oil, natural gas and electricity + and increased by VAT at 20%. For the post-2020 period, where the obligation coefficients are not yet defined, an + increase of 1%/year is considered. This assumption is equivalent to a constant amount of obligation assuming that + energy sales follow the trend decline of 1%/year. + +[^buildings]: https://www.ecologique-solidaire.gouv.fr/operations-standardisees + +```{eval-rst} +.. csv-table:: Cumulative kWh/m² per label transition + :name: cumac_label_transition + :file: table/policies/cumac_label_transition.csv + :header-rows: 1 + :stub-columns: 1 +``` + +```{eval-rst} +.. csv-table:: Amount in €/m² per label transition, for a CEE price of 4€/MWh cumac + :name: example_who_prices + :file: table/policies/example_who_prices.csv + :header-rows: 1 + :stub-columns: 1 +``` + +```{eval-rst} +.. csv-table:: Subsidy rate for a CEE price of 4€/MWh cumac + :name: example_who_subsidy_rate + :file: table/policies/example_who_subsidy_rate.csv + :header-rows: 1 + :stub-columns: 1 +``` + + +Subsidies under the WCO precariousness scheme are allocated according to the owner's income in the private sector and +according to the occupant's income in the social sector33. The taxes, on the other hand, are applied taxes are applied +uniformly to all households. + +The price of the WCO is the main determinant of the impact of the instrument. Conventional EWCs apply to households in +the C3 to C5 category and precarious EWCs apply to C1 and C2 households. The amount of the subsidy and the WCO produced +are doubled for C1 households to reflect the bonus granted to very modest households. The precariousness WCOs are +counted from 2015 (and not 2016 as in reality) in order to facilitate the comparison between periods 3 and 4. In line +with recent trends, it is assumed that the price of conventional and precariousness WCO is identical and capped at 20 +€/MWh cumac34. The price chronicle is subject to two scenario variants: + +- WCO scenario: 4€ from 2012 to 2016, 5€ in 2017 and then 2%/year increase from 2018 until the 20 €/MWh cumac cap. +- WCO+ scenario: 4€ from 2012 to 2016, 5€ in 2017 then 15€ in 2018, increasing by 2%/year from 2019 until the ceiling of + 20 €/MWh cumac. + +Since the proposed modeling only covers the residential building sector, two important looping mechanisms are missing +from a more complete evaluation of the instrument: the important looping mechanisms are missing from a more complete +evaluation of the instrument: + +- In theory, the WCO price should result from a market equilibrium and reflect the opportunity cost of the constraint + associated with the target. Taking these mechanisms into account implies representing all market actors, i.e. all + sectors covered by the scheme, including agriculture, industry and transport. Since the exercise proposed here focuses + on the residential sector, these adjustments cannot be taken into account. The price of WCO is therefore defined + exogenously in the residential sector, without any explicit link to the total target. +- In theory, the tax component of the WCO is defined by each obligated supplier in such a way that the revenue it + generates balances its subsidy expenses. To take this balance into account, it is again necessary to represent all + sectors, since an obligated fuel seller may, for example, pass on the cost of subsidies granted in the residential + sector to the price of fuel. In the exercise proposed here, the subsidies and taxes that apply to the residential + sector are The subsidies and taxes that apply to the residential sector are defined exogenously and independently. + + +## Reduced rate VAT (VATr) + +Renovation measures are subject to a reduced VAT rate of 5.5%, instead of the normal rate of 10% which applies in the +building sector. This assumption is embodied in our cost matrix. + +## Building code (BCO) + +Investment choices in new construction are limited to BBC and BEPOS levels between 2012 and 2019, then to zero energy +standard only from 2020. + diff --git a/docs/_sources/simulation_2012.md.txt b/docs/_sources/simulation_2012.md.txt index d5804b7..8968c46 100644 --- a/docs/_sources/simulation_2012.md.txt +++ b/docs/_sources/simulation_2012.md.txt @@ -1,5 +1,5 @@ -# Influence of input variables +# Simulation and sensitivity analysis ## Influence of exogenous variables diff --git a/docs/_sources/technical_documentation.md.txt b/docs/_sources/technical_documentation.md.txt index 4c46050..f8c1c98 100644 --- a/docs/_sources/technical_documentation.md.txt +++ b/docs/_sources/technical_documentation.md.txt @@ -176,7 +176,7 @@ into TWh can also explain the differences observed in wood consumption. ```{eval-rst} .. csv-table:: Calibration of total final actual energy consumption :name: calibration_energy - :file: table/ceren_energy_consumption_2012.csv + :file: table/input_2012/ceren_energy_consumption_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -259,7 +259,7 @@ by CGDD (2015) [^constructions] ```{eval-rst} .. csv-table:: Construction costs (€/m2) :name: cost_construction - :file: table/cost_construction_2012.csv + :file: table/input_2012/cost_construction_2012.csv :header-rows: 2 :stub-columns: 1 ``` @@ -277,7 +277,7 @@ abstract from short-term variations, the 2012 market shares are set in Res-IRF o ```{eval-rst} .. csv-table:: Distribution of heating fuels in new constructions in Res-IRF for year 2012 :name: heating_fuel_construction - :file: table/heating_fuel_construction.csv + :file: table/input_2012/heating_fuel_construction_2012.csv :header-rows: 1 :stub-columns: 1 ``` @@ -307,7 +307,7 @@ $INV_{i,f} < INV_{i,i+k} + INV_{i+k,f}$ for all k such that $1≤k + + + + + + + Modification from Res-IRF 3.0 — Res-IRF 0.0 documentation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    + + + +
    + + + + + +
    + +
    + + + + + + + + + + + + + + + + + + + +
    + + + + +
    +
    +
    +
    + + + +
    +

    Modification from Res-IRF 3.0

    +

    Modifications are listed in the table and the major changes are developed if necessary in a separate section.

    + + ++++++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    Table 40 Code modification

    Type of mistake

    Comments

    Status

    Date

    Bug

    Calibration of the market shares for construction was not considered in the code.

    Report

    Renovation rates shown in the report do not match the data used in the code

    Bug

    Renovation rates formula to calculate renovation rate objective was false.

    Report

    The discount rate displayed in the report does not correspond to the data used in Scilab (C4 collective housing = 5% and not 7% in the report).

    Report

    The calibration coefficients of the aggregated energy consumption (to match the CEREN data) displayed in the report do not correspond to the data used in Scilab.

    Bug

    ZIL doesnt’ target C building during the calibration year.

    Bug

    WHO tax have double count of the tax: once at initialization, again at the renovation stage

    +
    +

    Renovation rate function

    +

    The renovation rate \(τ_i\) of dwellings labelled i is then calculated as a logistic function of the NPV:

    +
    +\[τ_i=\frac{τ_{max}}{(1+(τ_{max}/τ_{min} -1) e^{-ρ(NPV_i- NPV_{min})})}\]
    +
      +

    • \(τ_{max}\) and NPV_{min} are constant based on own assumption,

      • +

    • \(NPV_i\) is a result from Res-IRF,

      • +

    • \(ρ\) is calibrated to replicate the observed renovation rates.

      • +
    +

    Previous work [Branger, Giraudet, Guivarch, and Quirion, 2015] showed that the model was very sensitive to this parameter.

    +

    Res-IRF 3.0, based on the Phébus building stock, replicated the renovation rates observed in the French housing stock +according to the Energy Performance Certificate (EPC) and the Decision Maker (DM).

    +

    Renovation rates by pair (ECD, decision maker) were determined using the following data:

    +
      +

    • 3% of all dwellings in the stock are renovated in the original year (source ADEME);

      • +

    • Share of renovations performed for each EPC Table 41: \(P(q | R)\)

      • +

    • Renovation rate by decision maker1 Table 42 : \(P(R | d)\)

      • +
    + + ++++ + + + + + + + + + + + + + + + + + + + + + + + + + +
    Table 41 Renovation share by energy performance certificate. Source: PUCA (2015)

    Initial label

    Contribution to the aggregate renovation rate

    G

    36%

    F

    30%

    E

    15%

    D

    10%

    C

    8%

    B

    1%

    + + +++++ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    Table 42 Renovation rate by decision make. Source : OPEN (2016) and USH (2017)

    Type of decision-maker

    Type of dwelling

    Renovation rate

    Owner-occupied

    Single-family

    4.70%

    Multi-family

    3.60%

    Privately rented

    Single-family

    2.00%

    Multi-family

    1.80%

    Social housing

    Single-family

    1.50%

    Multi-family

    2.00%

    +
    +

    Correction of renovation rate objective formula

    +

    The usual conditional probability formulas allow us to calculate rate_obj (renovation rate as a function of decision +maker and EPC) as follows:

    +
    +\[\text{rate_obj} = P(R | d \cap q) = \frac{P(R \cap d \cap q)}{P(d \cap q)} = \frac{P(R) P(d \cap q | R)}{P(d \cap q)}\]
    +

    Assuming independence between:

    +
    +\[P(d \cap q | R) = P(q|R) P(d | q \cap R) = P(q | R) P(d|R)\]
    +
    +\[\text{rate_obj} = P(R | d \cap q) = \frac{P(d) P(q) P(R|d) P(R|q)}{P(R) P(d \cap q)}\]
    +
      +

    • \(P(R)\) : aggregate renovation rate of the stock of 3%.

      • +

    • \(P(R|d)\) in Table 42

      • +

    • \(P(R|q) = P(R \cap q) / P(q) = P(R) * P(q|R) / P(q)\) and with \(P(q|R)\) in Table 41

      • +

    • \(P(q)\), \(P(d)\), \(P(d \cap q)\) calculated by the building stock structure.

      • +
    +

    Previously Res-IRF 3.0 calculated :

    +
    +\[\text{rate_obj} = \frac{P(R|d) P(R|q)}{P(R)}\]
    +

    We replace this formula by the formula defined above. We assess the marginal impact of this modification on Res-IRF +3.0. We observe a change in absolute value but not in trend.

    +
    +_images/flow_renovation.png +

    Fig. 23 Impact renovation rate formula on flow_renovation

    +
    +
    +_images/renovation_rate_decision_maker.png +

    Fig. 24 Impact renovation rate formula on decision-maker renovation rate

    +
    +
    +_images/renovation_rate_performance.png +

    Fig. 25 Impact renovation rate formula on EPC renovation rate

    +
    +
    +_images/consumption_actual.png +

    Fig. 26 Impact renovation rate formula on consumption actual

    +
    +
    +
    +

    Calibration segmentation

    +

    Parameter ρ is calibrated, for each type of decision-maker and each initial label, so that the NPVs calculated with the +subsidies in effect in 2012 (version 3.0) reproduce the observed renovation rates. In general, we don’t have enough +disaggregated data. +How can we calibrate an individual decision function per agent with only aggregated data?

    +

    A simple solution would be to say that each agent must replicate the observed renovation rates of his group. In this +case, we have as many decision functions as there are agents. This solution is not satisfactory because it erases all +the specificities of the agents (discount rate, investment horizon, etc…) and is equivalent to creating a model with a +representative agent. Also, a robust decision function must be unique for all agents.

    +
    +_images/function_agents.png +

    Fig. 27 Individual decision functions by agents.

    +
    +
    +

    Agents with the same observed renovation rate should get the same renovation rate function

    +

    Let say two agents got 2 NPV of 100 €/m2 (Agent 1) and 200 €/m2 (Agent 2) and their observed renovation rate is 3%. This +occurs when, due to lack of data, we assign two different agents the same renovation rate because they share common +attributes.

    +
      +

    • For example, Agent 1 and Agent 2 are both Homeowners in a Single-family dwelling, but one is living +in high-efficient dwelling when the other is living in a low-efficient dwelling. Agent 1 should get lower incentive to +invest than Agent 2.

      • +

    • Another example, will be between 2 agents of 2 different owner income class. Agent 1 owner could be +D1 with expensive investment loan when Agent 2 could become to D10.

      • +
    +
    +_images/agents_same_rate.png +

    Fig. 28 Individual decision functions by agents of the same group.

    +
    +

    These situations push us to create one renovation rate function for all agents sharing the same observed renovation rate, +this means the same attributes. The objective is to determine the parameters in such a way that the average weighted by +the number of agents reproduces the observed renovation rate.

    +

    The ideal solution would be to determine the function parameter by solving the following equation:

    +
    +\[\sum_{k=1}^n w(k) f(ρ, NPV_k) = \text{rate_obj}\]
    +

    However, this solution is not analytically solvable.

    +

    We are thinking of two other ways to calibrate the renovation function for all agents sharing the same attributes:

    +
      +

    • Calculate function parameter for each individual agent and then calculate an average parameter.

      • +

    • Calculate function parameter for a fictive individual agent that got an average utility function.

      • +
    +
    +_images/agents_comparison_rate.png +

    Fig. 29 Comparison of decision functions with the same renovation rate objective.

    +
    +
      +

    • There isn’t critical differences between the two methods.

      • +

    • There is no indication that the weighted average of the parameters allows us to recover the average value of the +observed renovation rate. This is an approximation.

      • +

    • Although without any theoretical support, we think it is simpler to imagine the average observed renovation rate +representing the decision of a representative agent.

      • +
    +

    We will therefore calibrate the decision function by the utility of a representative agent.

    +
    +
    +

    All agents should get the same renovation rate function

    +

    Should agents with different observed renovation rate get different renovation rate function?

    +
    +_images/agents_different_rate.png +

    Fig. 30 Should agents with different observed renovation rate get different renovation rate function?

    +
    +

    We believe that it is more robust to determine a single decision function for all agents. We thus define the decision +function as the best logistic function passing through the pairs (Utility, Investment rate) or (NPV, Renovation rate) +for our specific case.

    +

    Let say two agents with the following attributes: +Agent 1: NPV of 100 €/m2, and observed renovation rate of 2% +Agent 2: NPV of 200 €/m2, and observed renovation rate of 3%.

    +
    +_images/renovation_fitting_function.png +

    Fig. 31 Fitting renovation rate objective data to utility data calculated.

    +
    +
    +
    +

    Conclusion

    +

    We try to replicate the observed investment decision from a utility function per agent calculated by a model. However, +we do not have sufficiently disaggregated data to perform this calibration directly.

    +

    How can we calibrate an individual decision function per agent with only aggregated data?

    +
      +

    1. Define a representative agent by segmentation of the observed data,

      2. +

    2. Calculate the utility function for the representative agent,

      3. +

    3. Fitting observed renovation rate data to the utility calculated.

      4. +
    +

    Even without disaggregated data, each agent will have a different decision when calibrating. The average of the agents +per group of the same attributes will reflect the observed decision rates.

    +
    +
    +
    1
    +

    Parameters imported in the Scilab version of Res-IRF were different from those indicated in some articles. We +present here the correct values.

    +
    +
    +
    +
    +
    +
    + + +
    + +
    + +
    +
    + +
    + +
    + + + + + + + + + + + \ No newline at end of file diff --git a/docs/contributing.html b/docs/contributing.html index 8279fb8..b1f7134 100644 --- a/docs/contributing.html +++ b/docs/contributing.html @@ -43,7 +43,7 @@ - + @@ -94,9 +94,12 @@
  • License
  • Input Res-IRF version 3.0
  • Technical documentation
    • -
  • Influence of input variables
    • +
  • Simulation and sensitivity analysis
    • +
  • Setting up public policies
    • +
  • Policies assessment
  • Module
    • -
  • Additional information
    • +
  • Modification from Res-IRF 3.0
    • +
  • Installation help
  • Contributing
    • Updating Documentation
      • Overview
        • @@ -251,7 +254,7 @@

        Add content (adding .md file)