We  get  dependent  on  electronic  stuff. The  less  reliable  things  become  and we're  out  of  that's  an  view. Se  right  here,  is  an  assistant  for the  Seal  research  fellow at  the  Grant  of  Institute  at  LSE. And  he  received  his  PhD  in  2018  from the  school  Avance  study for  the  Social  Sciences  in  Paris. Aeron  dissertation, three  essays  on  Energy  prices  and the  energy  transition  received the  2019  award  for  best  doctor  of Dissertation  from  European  Association of  Environmental  and  Resource  economists. He's  currently  a  visitor at  the  Kennedy  School. And  you  put,  by  the  way, to  point  to the  essentially  the  French  equivalent of  the  counsel  of  economic advice  the  White  House. We  need  to  literally  to  the  White  House. Also  holds  that  MSC  and  for the  School  of  lines  of  Paris  Tech  and  MSC  and Economics  and  public  policy  for  C.  Is research  que  his  research focuses  on  the  interaction  between economic  inequality  and  climate  change mitigation  policies  in  order  to address  the  social  and political  acceptance  challenges that  hamper  implementation  of effective  decarbonization  topic  for today  is  from  re  innovation  to  green  jobs, please  join  me  in  it  paring  office. Hank  you. Thank  you.  Thank  you  so  much, Michael  for  this  very  nice  introduction, and  thank  you  so  much  for  inviting  me. I'm  delighted  to  be  here. So,  as  Michael  mentioned, this  project  is  really  about thinking  about  the  employment impacts  of  the  green  transition, but  from  the  standpoint  of green  innovation  impacts on  employment  outcomes. So,  this  is  joint  work with  my  colleague  Msatoz, also  assistant  professor  LC in  the  same  department, and  Francesco  Va, from  the  University  of  Milan. Um,  this  is  very  much  an  ongoing  project. Some  of  the  results  that  I  will  present to  you  today  are  fairly  preminary, so  please  keep  that  in  mind, and  I  really  welcome  your  feedback  as  we go  to  inform  this  project. So,  our  starting  point here  is  the  broader  question  of how  technological  transitions  affect employment  outcomes  and environmental  outcomes. So  there's  quite  a  bit  of  research  already about  the  impact  of automation  innovation,  unemployment  impacts, particularly  as  it  pertains to  the  impact  on  displacement  of  labor, with  lower  skilled  workers  being displaced  by  automation  innovation, and  also  on  its  effect  on  wage  growth, with  depressing  effect  on  wage  growth. And  also  the  it  tends  to polarize  the  job  market  with  a  lot  of new  jobs  creation  through automation  going  to  high  skilled  workers and  low  skilled  workers benefiting  less  from  that  type  of  innovation. But  there's  less  analysis  on the  impact  of  that  type  of innovation  on  green  innovation. Environmental  outcomes  and  particularly on  greenhouse  gas  emissions  reduction. So  that's  one  of the  things  we're  going  to  look  at  here. And  the  two  main  types  of technological  transition we're  going  to  look  at is  the  two  arguably  main  ones that  we're  going  through  globally. First,  of  course,  decarbonization. I  won't  have  to  motivate the  need  for  that  to  this  audience,  I'm  sure, where  we  see  across  the  developed  world, large  and  emerging  world, large  green  investments to  decarbonize  our  economies. And  one  ways  they  are  build  and  sold  to the  public  is  as massive  creators  of  employment. And  usually,  these  are  portrayed as  new  jobs  that would  be  relatively  high  skilled, so  quality  jobs,  that would  be  relatively  well  paid, and  that  would  be  numerous. So  one  of  the  things  that this  project  is  trying  to tackle  is  whether  we  can expect  that  to  be  the  case,  really. And  the  other  thing  that's  interesting to  us  is  whether  this push  for  greater  automation that  we  see  both  through development  in  AI  and broader  digitalization  of the  economy  can  have any  synergies  with  green  innovation, and  whether  it  can  actually help  hasten  the  reduction  of emissions  or  conversely  slow it  down  and  act  as  a  detriment  to  it. So  this  kind  of  idea  of a  twin  digital  and  green  transition happening  at  the  same  time  in  our  economies. So  just  to  make  a  bit  more  concrete, what  do  we  mean  by interactions  between  the  two, here  are  two  very  recent  examples, pull  out  from  the  press. On  an  example  here  on  the  left  hand  side of  a  solar  panel,  well, solar  power  plant manufacturing  company  really, that  is  developing  a  robot to  make  the  deployment  of solar  panels  into  power  plants both  faster  and  cheaper. That's  an  example  where automation  will  actually  help decarbonization  by  make  it more  accessible  from  a  cost  standpoint, And  here  on  the  right  hand  side, an  example  of  the  use  of modern  machine  learning  techniques  here  to help  make  the  grid  more  reliable  and  more amenable  to  higher  rate of  penetration  of  renewables. Because  as  you  know, when  you  have  higher  rates of  penetration  of renewable  energy  production, that  can  make  the  grid  unstable. And  here,  this  companies develop  is  using  tools developed  in  the  machine learning  world  to  help  stabilize  the  grid. So  again,  another  example  where  you  can  have positive  synergies  between the  digital  transition and  the  grid  transition. So  in  terms  of the  expected  employment  impacts of  the  green  transition. It's  not  just  used  as  kind  of a  crutch  by  policymakers to  sell  it  to  the  public. It's  also  expected  by  decision  makers  in the  private  sector  to  be a  big  source  of  new  jobs. So  here,  this  is a  survey  conducted  annually  by the  World  Economic  Forum  on  a  panel  of international  representative  decision  makers in  the  private  sector, and  they  asked  them  about  a  number, maybe  a  bit  small  to  see. But  essentially,  they  asked  them  to rank  a  number  of  global  trends  going  on. So  here  you  have  green  transitions, you  have  broadening  digital  access, you  have  the  COVID  pandemic, increased  geopolitical  tensions. So  all  the  main  trends and  asked  them  whether  they  thought that  it  would  be  mostly  a  source  of job  creation  or  mostly  a source  of  job  destruction. And  you  consider  the  ones that  consistently  ranked  the highest  are  the  green  trends. So  again,  there's this  really  broad  expectation that  the  green  transition  will be  a  net  source  of  job  creation, but  we  lack  empirical  evidence  for  that, and  that's  what  we're  trying  to  address  here. Another  example  of  that  is,  of  course, the  US  IRA,  which  was presented  to  the  public, mainly  as  a  investment  plan in  reindustrializing  America  and as  a  big  source  of  potential  job  creation. So  this  is  an  evaluation  run  by the  LEP  that  projected  that 1.5  million  jobs  could  be  created  by  the  IRA. But  again,  that's  predicated  on  the  idea  that renovation  will  not  be  a  net  job  destructor, and  that  would  not  act  as  automation, for  which  we  do  not  have  empirical  evidence, which  is  what  we're  trying  to  remedy. So  that  brings  me  to  the research  question we're  trying  to  address  here. The  first  one  is,  will emerging  technologies  like automation  and  green  innovation be  a  source  of job  creation  or  job  destruction? Well  the  compliment  workers or  substitute  for  workers. Second,  trying  to  better  understand how  innovations  in  green  technologies are  diffusing  through  the  economy. That's  one  really  challenging  thing  to  do, trying  to  find  reliable  empirical  measures of  technological  adoption  at  the  firm  level, and  that's  something  we're  trying  to  address. Third,  how  does  exposure  to  green  innovation. So  if  you  are  a  firm that  adopts  green  technology, what  does  that  mean  for  your  outcomes in  terms  of  employment? In  terms  of  the  types  of people  you're  going  to  recruit, in  terms  of  the  type  of wages  you're  going  to  provide, and  in  terms  of  environmental  outcome. So  dos  adoption  of green  innovation  lead  to enhanced  emissions  reduction  in  particular. And  finally,  whether  there's a  tension  between  being an  adopter  of  green  and  digital  innovation. So  the  idea  here  is that  adopting  digital  innovation, adopting  automation  driven  innovation is  meant  to  enhance  the  firm's  productivity. So  that  actually  run counter  to  the  objective  of  decarbonization. If  the  improvements  you experience  in  productivity  are going  faster  than  how fast  you  can  decarbonize. So  we  want  to  address whether  we  actually  see  any  evidence  for that  in  our  empirical  analysis. So  the  contributions  were trying  to  make  here  is  first, we  contribute  a  novel  measure of  technological  adoption  at  the  firm  level. So  here  the  idea  is  that we're  going  to  observe  the  content, the  occupational  content  at the  firm  level  through the  job  vacancies  they  post. So  essentially,  the  idea  is  that when  a  firm  wants  to  recruit  someone, it  describes  the  occupational content  of  the  position. And  that  description  contains the  type  of  skills  that  they  expect the  person  to  have  to  fill  the  position. And  those  skill  requirements actually  gives  us  information about  the  type  of  production  process and  the  type  of technologies  that  the  firm  uses. That's  really  like  the  kernel  of The  core  idea  of  our  measure. And  then  we  also contribute  what  we  believe  to  be the  first  link  between occupational  skills  and  patents. So  to  link  innovation with  the  occupational  content  of  work. Second,  we  try  to  answer the  question  of  whether g  innovation  is  labor  augmenting, meaning  whether  it  act  as a  compliment  for  worker, enhancing  their  capabilities  and potentially  acting  as  a net  positive  for  employment, or  whether  it's  labor  saving, which  is  acting  as a  substitute  for  workers  and acting  as  replacement  for  workers thereby  having a  negative  impact  on  employment. T  hird,  we  conduct a  causal  impact  assessment. So  here,  causal  understood  in econometrics  term  of  what being  an  adopter  of green  and  automation  innovation at  the  firm  level  means for  the  firm  level  outcomes, both  in  terms  of  employment  and  in environmental  terms  measured by  firm  level  emissions. And  finally,  y,  we  compliment  that  with the  environmental  aspect  of  things  through greenhouse  gas  emissions  observed at  the  establishment  level. So  I  know  that  like  some  of  you  will need  to  leave  at  1:00  P.M. So  I'll  skip  the literature  in  the  interest  of  time. We  feel  that  we  try  to contribute  to four  mainstreams  of  the  literature, just  to  go  over  them quickly  to  the  employment  impacts of  the  green  transition,  quite  obviously. We  also  contribute  to  how  technical  change affects  skills  requirements and  employment  impacts. So  the  broader  literature  on technological  transition  and  employment. We  also  contribute  this  new  measure  of firms  exposure  to innovation  and  technological  adoption. And  finally,  to  this  very new  and  emerging  literature on  how  this  idea  of twin  transition  between the  digital  and  green  transition  interacts and  may  deliver  benefits for  the  green  transition  itself. So  again,  knowing  that some  of  you  will  have  to live  in  half  an  hour, here's  a  preview  of  our  results. So  what  we  find  is  that the  expected  impact  of green  innovation  on labor  outcome  is  highly  heterogeneous. We  find  that  some  green  technologies are  expected  to  be  labor augmenting  and  actually as  labor  augmenting  as the  most  labor  augmented  technologies  we observe  outside the  world  of  green  innovation. And  other  green  innovation actually  at  the  opposite  end  of  the  spectrum, as  as  labor  saving  as technologies  we  observe elsewhere  in  the  economy. So  to  give  example,  we  find that  innovations  to  decarbonized information  and  communication  technologies  or to  green  buildings are  quite  labor  augmenting, and  I'll  explain  that in  a  minute  in  a  couple  of  slides. But  conversely,  innovations  in green  transportation,  particularly  in  EVs, or  in  Smart  grades  are  quite  labor  saving, and  thus  expected  to  have quite  detrimental  impacts  on  employment. We  also  find  that  looking at  a  dynamic  perspective, so  we'll  be  able  in our  sample  to  go  all  the  way  back  to  the  80s. And  so  over  this  like  four  decade  time  span from  the  80s  to  2020, we  find  that  green  innovation  has  become increasingly  less  labor  augmenting  over  time, which  again,  I  will  explain  in  a  minute. We  also  have  some  more  primary  results, p  empirical  analysis at  the  establishment  level. We  find  that  establishments that  are  early  adopters  of green  technologies  tend  to  create more  jobs  when  looking at  it  at  a  decadal  time  span. Essentially,  we  look  at  whether  you're an  early  adopter  of green  technology  in  the  early  2010, and  we  look  at  your  employment  outcomes in  the  early  2020s. And  what  we  find  is  that  the more  green  technology  we're  adopting, the  more  jobs  you're  creating. Now,  looking  at  the  nature  of  those  jobs, we  find  that  just  like  automation  before, reinnovation  tends  to  be slightly  skills  biased, meaning  that  those  jobs  tend  to  be more  prevalently  high  skilled  jobs than  low  skilled  jobs. But  the  extent  to  which  that's the  case  is  much  lower  than  for  automation. And  finally,  we  don't  find any  evidence  that  being  an  adopter  of automation  prevents  an  establishment or  a  firm  from  reducing  its  emissions. So  that's  actually  good  news  for kind  of  a  green  growth  type  of  story, whereby  you  can  still  have  productivity improvements  from  automation  and yet  still  be  able  to reduce  emissions  in  absolute  terms. All  right,  so  let's get  into  the  details  of  this. So  first,  the  conceptual framework  we  have  in  mind, driving  this  analysis. So  as  I  was  mentioning, we're  considering  two  types of  two  categories of  technological  innovation, green  innovation,  and  automation  innovation. And  we're  going  to  look  at their  impact  on  a  variety  of  dimensions, on  labor,  whether  it's a  net  job  creator  or  net  job  destroyer, on  whether  it  biases job  creations towards  high  skilled  workers  or  not, and  also  environmental  outcomes as  measured  by  greenhouse  gas  emissions. And  from  the  literature, we  have  some  priors  about  these. So  most  of  our  priors are  really  for  automation  because, as  I  was  saying,  that's  where  most  of the  research  has  been  happening  so  far. So  we  know  that adopting  automation  innovation  is probably  going  to  lead  to destructive  effects  on  labor, to  reduce  the  demand  for  labor. As  a  consequence,  or  sorry,  simultaneously, even  though  in  the  aggregate it  reduces  the  demand  for  labor, it  increases  the  demand for  high  skill  labors. It's  quite  scale  biased towards  high  skilled  workers. But  the  effect  on  emissions has  not  been  explored  so  far, so  that's  something  where we  don't  really  have  a strong  prior  and  where  we want  to  make  a  contribution. Conversely  on  green  innovation, the  one  where  we  have  a  strong  price  that hopefully  adopting  green  innovation  or adopting  green  technology  should make  you  better  at  reducing your  emissions  because  that's  the  whole  point of  green  technologies,  obviously. In  terms  of  its  effect  on  labor, we  don't  really  know  whether it's  labor  augmenting  or  labor  savings. That's  the  thing  that  we're  trying  to  answer. And  whether  it's  going  to  be  skill  biased, we  do  have  evidence  that green  jobs  that  have  been  created  so far  tend  to  be overrepresented  in  the  stem  professions. So  it  might  be  skills biased  towards  high  skilled  workers, but  again,  until  at  least  our  project, we  don't  think  that  we  had strong  evidence  for  that. And  in  terms  more  generally  on  the  impact of  labor  saving  versus labor  augmenting  innovation. So  this  is  a  taxonomy  to  classify innovation  that's  been  used in  the  labor  literature, particularly  to  look  at  automation  and AI  that  we  borrow  here. It's  effect  on  labor,  when  it's labor  saving  by  definition, it  is  destructive  for  jobs. And,  in  terms  of  wages, if  you  reduce  demand  for  labor, that's  going  to  depress  wages, going  to  make  workers  less able  to  bargain  for  higher  wages. Conversely,  labor  augmenting  innovation tends  to  have  a  positive  impact  on  labor, but  the  impact  on  wages  can  be ambiguous  because  it  will depend  on  workers'  ability  to take  advantage  of  those labor  augmenting  innovations. Remember, labor  augmenting  innovation  is  meant  to act  as  a  complement  to  workers'  skills, so  you  still  need  to  have  the  right  skills to  take  advantage  of those  labor  augmenting  innovations. Okay.  So  at  this  point, we  need  two  things  really to  answer  our  research  questions. First,  we  need  to  develop  that  novel  measure of  technological  adoption  at  the  firm  level. And  then  in  a  second  step, we're  going  to  need  to answer  that  question  of  whether green  innovation  is  labor augmenting  or  labor  saving. So  first,  how  are  we  going  to  go  about measuring  firm  level  technological  adoption? So  this  is  a  broader  challenge in  the  economics  literature,  as  I  was  saying, observing  which technologies  are  adopted  within the  firm  is  quite  tricky because  we  don't  have good  information  on  that. And  here,  for  our  design  to work  or  rather  to answer  our  research  question, we  need  even,  like  an  establishment  level. Adoption  of  technical  production. So  what  I  mean  by  establishment, think  of  it  as a  single  manufacturing  facility, a  single  manufacturing  site,  right? So  a  firm  will  be  composed of  multiple  establishments in  the  general  case  spread out  across  the  territory. So  usually,  one  way  to  go  about  this  is  to use  Peyton  counts  at  the  firm  level, so  to  look  at  the  firm, look  at  what  innovation produces,  what  Peyton  produces. Now,  we  don't  believe  that's  really  going  to work  here  because  particularly,  I  mean, that's  the  case  for  other  technologies,  too, but  particularly  in the  case  of  green  innovation, oftentimes  you're  adopting  innovation that  you  haven't  generated  yourselves. T,  for  example,  of  the  solar  industry, most  of  the  innovation today  is  happening  in  China, and  yet  you  have  firms  adopting  innovation produced  by  Chinese  firms  in the  solar  industry  all  across  the  world. So  in  those  case,  just  looking  at the  firm's  own  patent  counts  is  not  going  to cut  it  to  really  understand what  technologies  they're  adopting. So  here  as  I  was  mentioning  earlier, our  new  measure  is  going  to  build upon  observing  the  job  positions, job  postings  that  a  firm offers  to  understand  what is  the  scale  content  of  its  work. So  the  idea  is  that  skills that  firms  are  seeking  in their  job  ads  contain  information on  the  content  of the  work  that  people  are going  to  perform  in  that  firm. That  in  turn  will  tell  us  something  about the  type  of  technological  process they  use  in  their  production  process. And  so  that's  something  that has  already  been  explored  in  the  case  of  AI, particularly  with this  year's  Nobel  Prize  winner in  Economics,  Darn  Moglu. David  Alto  also  at  MIT, has  very  recent  and brilliant  work  on  that  topic. So  we  kind  of  adapt  that to  the  green  innovation  context. And  to  do  that, we'll  need  some  way  to  link  patents  that  are going  to  be  our  measure  of  innovation still  and  occupational  skills, which  is  where  we  get the  firm  level  observation of  whether the  technologies  are  being  adopted. So,  what  dataset  are  we're  going  to  use? First,  a  dataset  on  job  postings. So  we  take  advantage  of a  dataset  from  a  company  called  LCAst. If  there's  economists  among  you  might  have heard  of  it  as  the  Burning  Glass  Dataset. It  changed  name  recently. So  essentially,  it's  a  company that's  been  collecting  scraping, for  lack  of  a  better  word,  really, online  data  from  the  universe  of online  sources  for  job  postings  in  the  US, more  than  50,000  in  just  the  case  of  the  US, since  the  late  ts, our  data  set  here  is  going  to  start  in  2010. And  what  they  provide in  terms  of  guarantees  that  first, we  have  the  universe  of  all  online  job  ads in  the  entire  United  States. And  second,  it  provides structure  on  that  raw  textual  data. What  I  mean  by  structure  is  that  first, it  cleans  it  and  de  duplicates  it, which  is  already  like  a  massive  benefits. We  have  the  guarantee  that  each  job ad  we  observe  correspond to  exactly  one  position open  at  a  particular  firm. Second,  it's  going  to  extract structured  data from  the  raw  textural  description. The  way  we're  going  to observe  the  scale  content  of a  particular  position  is  not as  unstructured  natural  language  text, but  as  a  vector  of  skies. Right?  And  so  those  skills, they  come  from  a  taxonomy  that  LCs  has constructed  that  is  meant  to represent  that  universe  of  online  job  ads. So  32,000  unique  skills  that are  grouped  in  a  hierarchical  structure, and  that  provide  us with  information  on  that  skill  content. So  I'm  saying  linking skills  with  innovation,  skills  with  patents. That's  the  taxonomy  of skills  I'm  referring  to. And  we  also  observe other  the  dimension  of the  job  posting  offered, particular,  it's  occupational  category. So  here,  we  use  something  from the  Bureau  of  Labor  Statistics  called the  standard  occupational  classification, the  SOC,  which  is a  standard  classification  of the  whole  labor  force  into 700  distinct  occupations. We  know  the  wage  that  is  being  offered. We  know  the  educational  requirements. The  wage  unfortunately  has fairly  low  coverage  because  a  lot  of positions  are  being  advertised  without a  wage  range  being reported  directly  publicly. So  we  roughly  have  wage  in  25%  of  ads  only. We  also  know  the  company  name  with really  high  coverage  in  80%  of  cases, we  know  the  name  of  the  company that's  offering  the  position. And  we  also  know  the  location of  the  company  of  the  sorry, the  company  of  the  job, which  is  actually  very  useful. That's  how  we're  going to  define  an  establishment. So  there's  going  to  be a  combination  of  a  job  offered in  a  particular  location by  a  particular  firm. And  here  the  location, we  observe  it  at  the  county  level, at  least,  in  some  cases, we  observe  it  more  precisely, but  for  our  purposes, county  level  is  sufficient, because  the  pyramiter  that makes  sense  when  you're  looking  at labor  outcomes  is  something called  the  commuting  zone  in  the  US, which  is  essentially  the  size of  a  local  labor  market. So  here  establishment  is  going  to  be the  interaction  of  a company  in  a  commuting  zone. Some  limitations  about  that  dataset. Obviously, because  it's  an  online  job  ad  dataset, some  occupations  in the  high  scale  occupations can  be  over  represented. Although  this  has  gotten  better  over  time  as a  growing  portion  of  the  labor market  is  going  online. So  here  you  can  see  on  the  left hand  side,  I  dark  blue, it's  the  share  of  a  particular  occupation in  the  light  Cart  dataset. And  in  light  blue,  the  share  in employment  in the  actual  labor  force  in  the  US,  right? And  that's  the  average 2010-2024  that  we  use  here. So  you  see,  for  some  occupations, we're  very  close  to  being representative  of  the  actual  labor  force. For  others,  when  we  try  to have  like  nationally  aggregated  statistics, we'll  need  to  re  weight  them by  labor  force  weights. Another  thing  is  that  because we're  seeing  the  flow  of  new  positions, growing  firms  are  going  to  be over  represented  like  construction. So  that  means  that,  for  example, for  businesses  or  industries that  are  not  expanding, like  coal  mining,  if  you're thinking  about  energy  transition  issues, they're  going  to  be  underrepresented  in  that. One  thing  to  note,  though, despite  those  caveats  is that  the  dataset  is  so  large, you're  looking  at  330  million  job  ads 2010-2024, that  even  for  occupations that  are  underrepresented, we  still  have  a  lot  of  observation, we  still  have  quite  statistical  power to  conduct  analysis. Finally,  one  thing  to  really  keep  in  mind  is that  we  only  observe  the demand  side  of  the  labor  market. We  do  not  know  whether those  positions  were  filled, and  when  they  were  filled, we  do  not  know  if  the  position corresponded  exactly  to  what was  being  advertised  by  the  firm. So  that's  a  limitation  of that  approach  that  we  recognize. But  it's  also  still the  most  granular  observation you  can  get  with the  current  available  datasets  of the  skies  content  at  a  firm  level. So  that's  for  the  Labor  side. Now  for  the  Payton  side, we  use  Google's  Payton's  public  dataset. So  it  has  two  components. One  is  a  worldwide  bibiographic  dataset  on Peytons  that  has  global  coverage and  goes  the  way  back  to  the  19th  century. And  then  it  also  has  the  whole  US  PTO, US  patents  and  trademark  office, data  set  for  all  patents  text, title  and  abstract. So  here  we're  going  to  limit ourselves  to  title  and  abstract, and  we  also  observe the  classification  code  of  those  patents. So  Patents have  multiple  different  categorization. One  that  is  used  internationally  is the  cooperative  patent  classification or  CPC  classification. It's  a  hierarchical  taxonomy  that helps  tag  which  technologies particular  patent  relates  to. Single  patent  generally  will have  multiple  CPC  codes. I'll  get  to  in  a  minute to  how  we  use  that  in  our  context. So  one  thing  is  that  a  lot  of patents  are  actually  not necessarily  like  high  quality  patents, by  which  I  mean,  most  patents never  get  a  single  citation,  for  example. So  to  restrict our  datasets  to  quality  patents, we  define  a  certain  quality  threshold for  it  to  be  included  in  our  dataset. So  first,  we  require  that received  at  least  one  citation. And  second,  we  look at  where  that  patent  was  filed. So  in  general,  the  patent is  going  going  to  be  filed  in  one  country. But  if  a  patent  is  particularly  important, the  inventor  or  the  company  filing  it  will tend  to  file  it  in  multiple  locations. And  there's  one  particular  case where  a  patent  is  filed in the  three  main  patent  offices  in  the  world. The  European  patent  office, the  Japanese  patent  office,  and  the  USPTO. And  those  patents  are  called  triadic  patents, because  they're  find  those  three  locations. And  this  is  usually in  the  literature  used  as  like the  definition  of  like high  quality  patents  because it  takes  time  and resources  to  find  in those  three  of  these  locations. If  the  patent  was  worth  paying  that  cost, there's  probably  a  good  indication it  was  a  high  quality  patents. So  once  we  rest  we  apply  that  restriction, we're  left  with  1.6  million patents  1980-2019. And  that's  the  corpus  of  texts of  patent  text  that  we're  going  to use  in  our  analysis. So,  how  do  we  define  green  innovation? Here  we  rely  directly on  the  CPC  classification. So  the  CPC  has two  categories  that  have particular  relevance  to  us, O  is  so  called  YO  two  category, which  contains  all  technologies  relating to  cm  mitigation  and  adaptation, and  y4s,  which  is essentially  smart  grids  related  innovation. So  what  we  mean  by  green  innovation, this  paper  is very  much  climate  related  innovation, decarbonization  related  innovation. And  as  you  can  see, the  share  of  green  innovation, so  this  is  a  share  that  sedation  weighted 1980-2020  has  increased  quite  a  bit. Also  what  you  would  expect  from  P.  Now, just  zooming  in  in  terms  of  the  composition, the  main  nine  technologies  that  are identified  in  that  CPC  taxonomy  are  CCS, climate  adaptation,  green  buildings, decarbonizing  buildings,  greening  ICT, green  transportation,  mostly  EVs. Decarbonization  of  industrial  processes, renewables, smart  grids,  and  waste  management. And  as  you  can  see  in  terms of  composition  over  time, the  two  categories that  have  grown  the  most  are climate  adaptation  and  green  ICT. But  for  the  rest  of  the  technologies, the  split  is  fairly  stable  over  time. Now,  turning  to  automation. Here  we  rely  on  a  recent  contribution by  our  colleagues, David  Demus  Antoine  Sept,  Martin  Olson, and  Cars  Asana,  forthcoming  in the  JPE  who  classify patents  as  being  related  to  automation. And  for  that,  they  use a  keyword  based  approach that  that's  fairly  detailed, and  in  which  they  focus on  machinery  innovation. So  we're  kind  enough  to  share  with us  the  CPC  codes they've  identified  as  being automation  relevant  at  the  six  digit  level. And  so  here,  since  we're looking  really  at  automation  with  imaginary, the  way  to  think  about  how  we  define automation  is  in  terms of  automation  of  the  production  process, industrial  production  process,  so  to  speak. So  here  you  can  see  this  kind  of  like  in U  shape  pattern  over  time  in  the  share of  automation  patents  since  the  80s. And  here  that  spike  is  really related  to  innovation  in  industrial  robots, whereas  here  in  the  more  recent  it's  more about  digital  automation  in  the  2010. All  right.  So  now,  how  are  we going  to  link  those  two  datasets? How  are  we  going  to  link  innovation to  scale  content,  patents  to  skills. So  here  we're  going  to  use modern  NLP  techniques. More  specifically,  sentence  transformers. I'm  not  going  to  have  time  to  get into  details  of  how  to  implement  that. But  the  gist  of  it  is  that if  you  have  two  textural  sources, which  we  do  here,  we have  patent  titles  and  abstracts, so  that  describes  the  content  of  a  patent. And  for  each  skill,  we  also  have a  paragraph describing  the  content  of  that  skill. What  sentence  transformers will  allow  us  to  do  is  to compute  a  score  of how  semantically  similar  they  are, so  how  semantically  related  they  are. And  that  semantic  similarity is  how  we're  going  to  construct the  link,  essentially. So  for  each  of  the  1.6  million  patents, we  calculate  a  semantic  similarity  score with  each  of  the  32,000  skills. So  you  can  think  of  it  as like  a  really  large  matrix with  1.6  million  rows  and  32,000  columns. Then  we're  going  to  aggregate those  patents  at  the  four  digit  CPC  level, which  is  going  to  let  us  have our  nine  green  technologies split  out  here  because they're  at  the  four  digit  level, and  another  eight  for  automation. And  then  we're  going  to  define  a  threshold. So,  I  don't  know  if  I  have  a  mouse here  just  to  give  you  one  example. So  what  we  get  at  that  point  is  for each  of  these  four  digit technological  categories, a  distribution  of  semantic  similarity, so  we  normalize  it  0-1. So  each  of  these  is  a  skill. So  we  have  the  distribution  of  the 32,000  skills  ranked by  high  semantically  similar. They  are  to  each  of those  technological  categories. And  we're  going  to  choose  a  threshold  here, and  everything  that's  above  that  threshold, we're  going  to  consider  that  skills that  are  above  the  threshold are  characteristic of  that  particular  technology. So  we're  going  to  call  them  marker  skills. So  here  this  is  what  you  see. Go  back  over  here. This  is  what  you  see  here. For  example,  storm  drains. That's  a  marker  scale for  adaptation,  climate  change  adaptation. So  the  idea  is  that  when  we  observe  a  job posting  that  requires  the  scale  storm  drains, then  we  know  that  that  firm  will  stand  to benefit  from  innovation  in adaptation  because  of  that  link. Similarly  here,  y4s  remembers the  smart  git  grid,  s  technological  category. On  market  scale  for  it  is smart  meter  systems. Here,  B  25  data  that's  actuators, A  and  market  scale would  be  compuon  umerical  control. So  moving  to  our  job  dataset, and  the  colors  are  wrong,  but  that's  fine. This  is  supposed  to  be  green. So  essentially,  here, these  are  real  examples taken  from  the  dataset. For  this  flood  safety  engineer requires  a  skill  storm  drains. So  we  know  from  that  that in  that  particular  firm, innovation  in  storm  drain related  technology  in  Y  02  A, the  climate  chain  adaptation  category, will  stand  to  benefit  that  particular  firm. Here,  this  automation  technician has  a  CNC  skill  that's  required. So  the  firm  that  employs  them  will stand  to  benefit  from  automation  innovation in  the  CNC  category. But  of  course,  most of  the  ads  we  observe  will  have neither  a  green  marker  skill nor  an  automation  market  skill. So  it  will  just  be unclassified  as  it  relates to  our  two  technological  classes. Now,  when  we  look  at  the  share, so  our  metric  really is  for  each  establishment  or  each  firm, what  is  the  share  of  ads  that  contain a  green  market  skill or  an  automation  market  skill? And  that's  how  we're  going  to  define our  technological  adoption, technological  exposure  metric. So  looking  at  it  for  green, It's  a  big  crushed  by the  scale  of  automation, but  there's  still  like  an  uptake of  a  time  from  2010  onward, from  4%  to  5%. So  these  are  not  re  weighted. So  this  is  like  directly taken  from  the  LCAs  job  ad, bear  in  mind  that  re  weighted by  employment  shares, that  the  picture  might  be  a  bit  different. But  that's  dwarfed  obviously by  the  growth  in  exposure  to  automation. So  the  share  of  ads, highlighting  adoption of  automation  technology really  shot  up  over  the  past  15  years. Now,  zooming  in  on  green  technologies, we  see  that  green  ICT  is  the  biggest  share here  and  that  you  have  an  uptake  in smuggle  adoption  and industrial  process  decarbonization. And  you  see  this  like  a  slight  increase 4-5%  over  the  period. All  right.  So  that's  for  the  first, like  constructing  a  measure  of technical  adoption  at  the  firm  level. Now,  what  about  how  to  identify  whether renovation  in  general  is labor  saving  or  labor  augmenting. So  here  we  borrow  from David  Otters  recent  paper  in  the  QG, whereby  they  propose  to operationalize  that  concept  of being  labor  saving  or  labor  augmenting. And  the  way  they propose  to  do  that  is  the  following. So  we  know  that  labor  saving  innovation  is meant  to  replace  tasks that  workers  already  conduct. So  the  idea  they  introduce  is  to  look at  the  textural  content  of  a  patent and  look  at  whether it's  symetically  similar  to existing  tasks  that  we already  observed  from  datasets produced  by  the  BLS. So  essentially,  the  idea is  that  if  a  patent  is symetically  similar  to  existing  work  content, then  it  stands  to  replace workers  in  more  likely  fashion. Conversely,  for  laboring  menting  innovation, it's  meant  to  complement  workers. So  the  idea  here  is  to  look at  you  take  a  period,  for  example,  a  decade. You  look  at  the  job  titles  that  will appear  in  the  US  Census at  the  end  of  that  decade. And  if  a  Paton  fight  before that  relates  to  future  job  titles, Then  the  idea  is  that  because labor  augmenting  innovation  is  meant to  increase  the  number  of occupation  and  create  new  jobs and  new  types  of  occupation, then  if  a  patent  relates to  future  job  titles, then  it's  tend  to  be  labor  augmenting. And  we  already  have  all  the  data  we need  to  implement  the  mythology. With  our  centers  transformer  approach, which  is  not  what  we're  using, and  apply  it  to  green  in  technology  classes, which  also  can't  do  in  a  dataset  to  look at  what  we  observe  for  green  novation. And  what  we  observe  what  we  obtain  is  this. There's  a  bit  to  impact  here. So  on  the  x  axis, essentially  what  you  have  is  the  ratio  of labor  augmenting  patents to  labor  saving  patents. So  the  further  to  the  right  you  are, the  more  labor  augmenting  you  are, the  further  to  the  left  you  are, the  more  labor  saving  you  are. Okay? Here  in  blue,  you  see broad  technological  classes  that cover  everything  except for  green  technologies. The  analysis  is  done  at  the four  digit  CPC  level, at  the  technology  class  level, which  is  why  you  have  a  distribution for  those  broad  technology  classes, we  only  have  one  observation for  each  of  the  green  technologies. And  what  this  tells  us  is  the  following. First,  that  as  I  was  mentioning  earlier, green  ICT  and  green  buildings, they  tend  to  be  quite  labor  augmenting, actually  compared  to  the  rest  of the  technological  classes  tend  to be  amongst  the  most  labor  augmenting. It  makes  sense  because  if  you  look  at the  top  two  labor  augmenting broad  technology  classes, ICT  is  very  close  to  instrument  information, which  is  the  ICT  technology  class, and  Green  Buildings  is  a  lot about  sensors  and  electronics, which  is  also  very related  to  electricity  and  electronics. So  the  fact  that  these  these  two  stand together  is  what  we  would expect  and  it  is  confirmed  by  this  nysis. Fversely,  if  you  look  at  green  transportation and  compare  it  to  broader  transportation, green  transportation  tends  to  be a  most  the  most  labor  saving  type of  transportation,  really. And  again,  that  we  would expect  because  we  know  that  producing, manufacturing  an  electric  vehicle requires  a  lot  less  labor than  manufacturing  an  EV  IC  vehicle,  sorry. And  so  what  this  tells  us  is  that innovation  in  the  EV  space tends  to  even  reinforce  that  trend. Make  it  increasingly  labor saving,  so  increasingly  automated, so  to  speak,  to produce  decarbonized  vehicles, electric  vehicles. Here  you  see  what  I  was  mentioning  earlier that  amongst  the  most  labor  saving, you  have  smart  grids  and  CCS,  which  again, are  not  very  labor  intensive  technologies. So  again,  as  expected. Also  compute  those  metrics  over  time, and  that's  where  we  find  what  I was  mentioning  in  the  results  preview. Here,  we  only  look  at the  nine  green  technological  categories. And  when  we  look  at  it  from  the  80s  onward, we  find  that  earlier  on, those  green  innovation  green  patents tended  to  be  a  lot  more  labor  augmenting, and  as  we  move  onward, they  tend  to  become  more  labor  saving. So  one  way  to  interpret that  that  makes  sense  is that  if  you  think of  it  in  terms  of  an  innovation  cycle, earlier  on,  you  kind  of  at the  design  stage  of  the  energy  transition, at  the  design  stage  of  decarbonization. So  you're  really  not considering  implementation  yet. So  at  that  point,  it  makes a  lot  of  sense  for  it  to  be  labor  augmentic, because  it's  meant  to  it's  going  to create  a  lot  more  new  occupations, a  lot  more  work  because  you're  designing  it, and  it's  going  to  be  high  scale  occupations. Whereas,  as  we  move  towards  the  2010s, we  have  an  increased  need  for implementing  the carbonization  and  the  energy  adicion. And  at  that  point,  it  makes more  sense  to  focus  more  on labor  saving  innovation, which  are  more  needed  when  you  get  at the  implementation  stage  at the  R&D  design  stage. So  now  we  have  all  of  that. We're  ready  to  answer  our original  question,  which  was, what  is  the  impact  of green  an  atomation  innovation on  labor  emissions? So  the  metric  I've introduced  the  novel  metric of  technology  adoption. We  can  actually  compute  it over  our  time  frame,  2010-2024, we  can  observe  it  dynamically  over  time, and  we  can  aggregate  it at  various  levels  geographically. We  could  look  at  it  at commuting  zone  level  for regional  kind  of  impact. We  could  compute  it  at  the  firm  level, but  we  could  also  look  at  it at  the  establishment  level because  all  we  need  is  adds at  the  establishment  level. And  so  that  allows  us To  create  an  empirical  design to  assess  the  causal  impact  of an  increase  in  innovation  in green  technologies  and automation  technologies on  labor  and  environmental  outcomes. So  what  we're  going  to  consider  as I  was  saying  is  long  differences. So  compared  the  early  2010s with  the  early  20  tents. And  the  reason  we  do  that is  that  we  know  that technological  diffusion is  not  a  fast  process. So  you  need  to  allow  for  at delay  to  re  observe  impacts. And  finally,  given  the  way we  define  our  metric of  technological  adoption, there  would  obviously  be a  risk  of  endogenity, a  risk  of  reverse  causality. So  the  way  we  do  this  is that  we're  going  to  design  an  instrument for  that  change  in  technological  adoption 2010-20  tents, and  that  instrument  essentially  going to  consider  that  what motivates  firm  to  adopt more  green  technologies  or more  automation  technologies  is going  to  be  driven by  innovation in  those  technologies  outside  the  firm. So  I'm  just  going  to  describe that  more  specifically. So  just  to  have a  common  ground  of  what  we're  talking  about, The  way  we  can  write  our  metric of  technological  adoption  is  the  following. Consider  an  establishment  I, consider  the  set  of  ads offered  by  that  establishment I  in  a  given  period. And  here  this  indicator  function is  going  to  be  whether  a  given  add A  contains  a  market  scales for  technology  K.  So that  share  of  ads  is  just  a  share  of  ads  at the  establishment  level  that pertains  to  a  certain  technology. So  that's  going  to  give  us  shares  for each  of  the  four  digit technologies  that  we've  identified. We  can  aggregate  them  then  into our  two  main  technological  categories  of interest,  green,  and  automation. So  we  have  those  shares  for each  establishment  over  time. Now,  The  shift  share that  I  was  referring  to  is  that  we're going  to  look  at  the  change  in technology  adoption between  the  2010s  and  20  tents. So  we're  going  to  look at  the  change  in  that  share at  the  establishment  level for  green  and  for  automation. As  I  was  saying,  this  risks being  very  endogeneous. So  we  need  to  instrument  for a  plausibly  exogenous  instrument. The  instrument  we  choose  is  to use  a  shift  share  type  of  instruments. I'm  not  going  to  get  to  too  much  detail, but  essentially  the  idea  is  to  look at  the  level  of adoption  in  the  baseline  period, in  the  early  2010  for each  of  the  technologies  and use  an  exogenous  shifter that  could  have  driven the  change  in  adoption. And  here,  the  exhoit  shifter  we're  going  to use  is  innovation  in each  of  these  technologies. So  in  particular,  We  also want  that  innovation  to  not  be  endogeneous. So  to  make  sure,  since  we're only  looking  at  US  based  firms,  right, to  make  sure  that  it's  innovation happening  outside  of  the  firm, not  only  are  we  going  to  consider innovation  happen  outside  the  firm, we're  also  going  to  consider  innovation happens  outside  of  the  US  entirely. So  we're  no  longer  going  to  look  at  patents that  are  triadic  for this  particular  instrument. We're  going  to  look  at  so  called diadic  patents  that  are  only  find  in two  paid  offices  in  the  EPO  and  the  JPO. And  that's  what  we're  going  to  use to  construct  our  instruments. Okay. So  in  terms  of our  definition,  as  I  was  saying, the  establishment  is  going  to  be a  firm  advertising  within  a  commeting  zone. The  baseline  period  that  we're  going  to use  is  2010,  2014, so  that's  to  get  kind  of  a  sufficient  sample in each  establishment  to  construct  our  shares. The  outcome  period  is  going  to  be 21  to  2021  to  2023. And  the  volume  of  innovation, the  flow  of  innovation to  construct  the  instrument, we're  going  to  consider  the  flow  innovation 2005-2015. So  here  we  consider  that  effect  that  period. So  implicitly,  the  hypothesis we're  making  is  that  we  allow  for kind  of  a  six  year  delay for  technical  adoption  from the  Patent  publication  to  observing effects  on  labor  environmental  outcomes. And  here  in  this  long  difference, the  types  of  outcome  we're  going  to look  at  are  employment, skills  ratio,  wages,  and  emissions. So  I  just  have  time  to present  the  results,  so  that's  great. Starting  with  number  of  jobs. So  here,  The  dependent  variable is  the  difference  in  the  number  of job  ads  offered  at  the  establishment  level 2021-2023  and  the  baseline  period of  2010  to  2014. Okay.  So  that's  this  change  in  D  log. So  the  way  to  read  those  coefficients  is that  a  1%  change  in the  adoption  metric  of green  technology  drives  a 0.4%  change  in  the  number of  positions  offered,  right? So  what  we  find  is  that green  technology  adoption drives  employment  creation. That  establishments  that  adopt more  green  technology  create  more  jobs. And  again,  we  claim  that  this  is a  causal  relationship  because of  our  empirical  design. Conversely,  as  we  would expect  from  the  literature, adoption  of  automation  technology is  actually  a  net  destructor  of  jobs, like  a  1%  increase in  our  technological  adoption  metric  for automation  drives  almost  a  1%  reduction in  the  number  of  jobs  offered. Now,  turning  to  the  ratio of  low  and  high  skilled  workers. So  here,  what  we  look  at  is  the  ratio  between the  number  of  high  skilled  positions offered  to  the  number of  low  skill  position  offered. So  an  increase  in  that  ratio  will  mean  that the  technological  change  we're  looking  at  is skills  bias  towards  high  skilled  workers, because  it  tends  to  create  more  jobs for  high  skilled  workers than  low  skilled  workers. And  here  what  we  find,  as expected  again  from  the  literature, automation  technological  adoption  is bias  towards  high  skilled  workers. For  green  innovation,  we also  find  when  we  look  at  it  by  itself, it's  less  physically  significant  where  we combine  it  with  automation  exposure, but  also  the  magnitude  is half  that  of  automation  adoption. So  green  innovation  seems  to  have also  the  same  characteristy  of being  biased  towards  high  skilled  workers, but  to  a  lesser  extent  to a  lower  extent  than  automation. So  the  fears  that  it  would polarize  the  job  market  towards higher  skilled  workers  that is  definitely  warranted  for automation  technological  adoption  is less  warranted  for  green  innovation. Now,  looking  at  wages,  again, with  all  the  caveats that  I  mentioned  earlier, we  only  observed  wages  in  25%  of  our  sample. And  in  addition  to  that, this  is  the  offered  salary. We  don't  know  if  it  was  actually the  salary  taken  by by  the  person  filling  the  position. But  what  we  do  find,  which  is not  exactly  great  news,  I  suppose, is  that  both  green and  automation  innovation  tend  to  depress wages  or  green  and  automation  adoption. Automation  that  was expected  from  the  literature, green,  that's  a  new  contribution,  and  it's. It's  more  moderate  again  than  automation, but  it's  still  a  pretty  strong  effect. And  finally,  our  last  result  is looking  at  CO  two  emissions at  the  establishment  level. So  one  caveat  about  that  again  is  that the  source  of  our emissions  data  is  from  the  EPA, the  EPA  has  a  large  emitter  plants  dataset. So  we  had  to  manually merge  that  with  our  Job  data, which,  as  you  can  imagine, it  was  a  lot  of  fun. The  problem  is  there's  very few  establishments that  remain  in  that  dataset because  the  constraints  are we  need  to  have  a  match  between the  EPA  dataset  and  Jobs  dataset, and  we  need  to  observe  it  in the  early  2010s  and  the  early  2020. When  you  have all  those  restrictions  in  place, you're  down  to  less  than  fewer than  1,000  observations, fortunately,  few  1,000  establishments. So  we  don't  achieve stistical  significance  here. But  in  terms  of  the  signs, so  we  have  the  expected  sign. You  have  a  reduction  in emission  the  more  you  adopt  green  technology. But  reassuringly, we  don't  observe  that  increased  adoption  of automation  technology  would  drive a  growth  in  emissions. So  we  can  reject  the  idea  that  increased automation  technology  adoption  might be  a  detriment  to  emissions  reduction, and  that's  actually  quite  reassuring  for the  green  growth  story that  was  mentioning  earlier. So  just  in  conclusion to  recapitulate  what  we've  been  over, Gnovations  impact  is  expected  to be  highly heterogeneous,  depending  on  technology. H  technologies  are  going to  be  quite  labor  saving. Others  are  going  to be  quite  labor  augmenting. However,  as  a  broad  category, it  has  become  less and  less  labor  augmenting  over  time. And  we  also  have  preliminary  results that  suggest  that  increases in  adoption  of  green  technologies  at the  firm  and  establishment  level  drives  jobs, creations. We  also  have  some  suggestive  evidence  that this  job  creation  is  going  to  be a  skills  biased  towards  high  skilled  workers, but  to  moderate  extent, certainly  to  a  lower  extent than  for  automation  innovation. And  reassuringly,  we  don't  find any  evidence  that  increased  adoption of  automation  technologies  might weaken  emissions  reduction  moving  forward. Just  to  give  you  a  kind of  a  preview  of  what we're  working  on  right  now. So  we  exploring  the  heterogenity  of those  findings  of  our  firm  sizes. We  already  have  evidence that  a  lot  of  automations, for  example,  job  destruction  potential is  actually  concentrated  in  larger  firms. We  don't  find  any  evidence of  it  in  smaller  firms. We  also  looking  at  hetrogeny across  industries  across  occupations, whether  what  the  Litur  said  about  how stem  professions  tend  to  benefit  from the  green  transition  is verified  in  our  design. And  finally,  we  also  want  to  look  at spillover  effects  at  the  more  regional  level at  the  commuting  zone  level  in  particular. The  idea  being  that  here, because  our  design  requires that  you  observe  the  firm  booth  in 2010  in  the  early  2010  and  early  2020s, we  can't  look  at  firm  entry. And  since  green  technology has  been  growing  a  lot  over  that  period, we  also  want  to  know  whether having  regional  specialization on  green  adoption  of technology  also  yields  benefits in  terms  of  firm  creation, the  same  space  over that  same  decade  all  time  span. That's  it. Thank  you  very  much.  Take  questions. Absolutely.  So  the  bubble in  AI  electricity  demand is  after  the  end  of  your  study. Yeah.  You  can't  say  anything about  the  reality  of the  importance  of  indsion? Absolutely.  So  it's  a  big  concern, indeed  that  increased  in power  demand  from  AI would  change  those  findings. So  one  thing  is  that that  design  looks  at automation  adoption  within  the  firm. AI  power  demand  is  mostly  concentrated  in data  centers  that  are  mostly concentrated  in  like  very  few  firms. So  actually,  that  would not  be  reflected  in  the manufacturing  firms  emission  that is  what  we're  observing  here. So  we  wouldn't  I  mean, Uh,  that's  something  I  didn't  mention, but  the  sector  we're  limiting  ourselves to  is  mining,  utilities, and  manufacturing  here,  because  those  are the  firms  that  appear  in the  EPA  High  Mitter  dataset. So  we  would  to  observe  that. Um  In  terms  of  the  magnitude  of  it, I  know  big  tech  companies  are  pretty big  on  moving  towards  the  carbonizs  of  power, so  that  they're  going  to reopen  three  mile  island. Microsoft  stroke  a  deal  to  do  that. They're  definitely  locating  data  centers  in areas  where  there's  a  lot  of availability  of  clean  power. So  the  exact  magnitude of  that  still  remains  to  be  seen, but  in  terms  of  observing  it, it  will  be  super concentrated  in  very  few  firms. So,  I  think  in  you  would  only  see  it  in really  scope  three  emissions and  not  really  in scope  one  and  two  emissions. Pick  it  up  a  qual  Essentially. Okay.  We  are  a  little  late. Let  me  suggest  we  get  five  minute  for  sure. Thank  you  very  much. My  a  very  general  topic  was  interested now  on  your  view you  were  watching  the  energy transition  for  many  years  a. So  what's  your  view  on the  impact  of  change  in government  or  feeding  of. Let's  start  with  an  easy  and non  polemic  question. So  I  mean,  we  were just  talking  about  that a  couple  of  hours  ago. I  guess,  in  the  case  of  the  US, like  the  landmark  climate  related  policy, we  specifically  designed  to  be  as administration  change  proof  as  possible  with a  lot  of  funding  going  to red  districts,  essentially. So  that  would  hopefully  make  it  very hard  for  the  incoming administration  to  repeal  it. Now,  what  I'm  more  concerned about  is  the  growing  evidence  we have  back  home  in  Europe of  a  growing  backlash  towards  climate  policy. So  here  in  Europe,  it's  not  a  case that  people  don't  believe  that  clime  is  real. It's  not  even  the  case  that  they don't  support  climate  action, is  that  when  they  see  what  climate  action actually  means  for  the, you  know,  the  purchasing  power, their  day  to  day  life,  then they  start  to  balk  a  bit  at  it. And  that  part  thing we  really  haven't  solved  it. And  that's  one  of  the  driving motivation  of  that  paper,  too, that  if  we  keep  promising  that large  scale  investment  in climate  change  will  deliver  a lot  of  high  quality, well  paying  jobs,  and then  it  fails  to  deliver  that. The  pushback  will  be  massive, and  that  ultimately  will delay  climate  action  massively. So  I  think  to  me  that the  main  concern  is  rather  that pushback  on  com  policy  may be  strong  enough  to  drive  income, you  know,  new  administrations that  would  actually  be  anti  Cmin  action, but  that  it  would  actually  come  from the  population  rather than  from  the  administration, if  you  see  what  I  mean? I  was  about  to  mention,  yeah. I  mean,  what  happened  with. Yes,  thanks. I  have  a  question  specifically about  transportation  industry. What  you  think  vehicles on  the  automation  side  equal So  actually,  I  mean,  I  haven't seen  that  much  research  on  that. I  think  it  would  be a  very  interesting  question  to  explore. What  little  I've  seen is  that  as  we  know  utilization  rates of  current  vehicles  are super  low  because  they're privately  owned  and  they  stand still  for  90%  at  the  time. So  the  idea  is  that  with  autonomous  vehicles, you  could  move  to  having  vehicle  fleets that  would  have a  much  higher  utilization  rate. And  from  the  aggregate  perspective, the  resource  intensity  of delivering private  transportation  would  be  lowered. Whether  that  actually  bears  out. I  mean,  the  same  argument  had  been  made about  UBA  15  years  ago  and assessments  that  have  been carried  out  since  then revealed  that  it  actually  just  has increased  demand  for  private  transportation, has  increased  aggregate  emsions, where  we  could  assess  it. So  I'm  not  sure  we actually  delivered  that  promise, but  that's  what  proponents of  that  approach  are  advocating  at  least. That's  a  great  question  that actually  relates  to  two  papers. Thanks  On  the  first  aspect, most  of  the  job  that  is being  created  right  now  or  most  of  the  jobs that  are  recognizably  green  jobs actually  are  towards  the  lower  end of  the  skies  distribution. So  if  you  look  at  within  occupation, an  installer  that  does green  installations  will  tend to  be  more  skilled. Than  a  non  green  installer. One  way  to  think  about  this,  if you  think  about  heat  pump  installations, a  plumber  that  can  also  install heat  pumps  needs  extra  qualifications to  be  able  to  install  heat  pump. That's  what  I  mean  by  being  more skilled  with  an  occupation. But  if  you  look  at  where growth  and  green  jobs  is  happening, it's  mostly  amongst  low  skilled  workers, because  what  we  need  to  make  decarbonization happen  is  actually  a  lot  more  installations, technicians, maintenance  technicians, construction  workers, rather  than  engineers  and  sciences. I  mean,  we  also  need  them,  right? But  for  the  actual  day  to  day  work  of implementing  the  decarbonization, that's  not  what  you  need. So  completely  agree  with  you  on  that  point. However,  walking  back  to the  earlier  point  about  pushback, if  you  don't  take  into  account whether  energy  transition  will affect  spatial  inequalities, then  you're  in  for  a  massive  pushback. In  particular,  you  mentioned  the  rust  belt, that's  exactly  an  example where  that  was  not  well  handled. And  if  you  concentrate  economic  difficulties, particularly,  here  we're  looking really  on the  other  the  green  jobs  aspects in  this  paper. But  if  you  think  on  the  flip side  of  that  story  about foss  intensive  jobs  that  are  going  to disappear  in  the  decarbonization  story,  well, What  we  find  in  another  of our  papers  is  that those  false  intensive  jobs, particularly  in  the  US,  tend  to  be the  last  remaining  well  paying  jobs in  relatively  impoverished  communities. The  posted  chart  for  this  would  be  like Northern  Dakota  with  the  B  in jail  or  like  the  Appalachians. If  you  remove  those  mining  jobs from  those  communities, they  have  nothing  left. And  the  fact  that  you're  creating, solar  PV  solar  jobs  in California  is  not  going to  do  them  a  whole  lot  of  good. So  being careful  about  that  and  about  this  impact, I  think  is  crucial  to  avoid  the  kind  of pushback  that  would  set  us back  ten  or  15  years. So  first  ask.