 Okay, so hello and welcome back to the LFS and APS user conference. I really hope you enjoyed this morning's session and the presentations. If you've been here since the beginning. I now have the pleasure to introduce three more presentations in this session. Just a reminder for all attendees, you are now in session 3A. I hope that's the one you were after. Otherwise, you would have to join the other link for session 3B. So we've got about half an hour slot for each presentation, including about 20 minute presentations and some time for questions after. I would ask attendees to put their questions in the Q&A chat at the bottom and I can read them out to the presenters afterwards and we can go through them one by one. And if we haven't got enough time for questions, then presenters will be able to answer them directly in the chat afterwards. Right, first one up is Andrew Bryce from the University of Sheffield, talking about the segregation, segmentation and disability gaps in the labour market during COVID-19. Andrew is a research associate in the Department of Economics at the University of Sheffield and he and his co-authors were funded by the Health Foundation to investigate the labour market impacts of health in the UK and they have currently just started a three-year project unpacking the disability and employment gap in the UK funded by the Newfield Foundation. Okay, Andrew, if you're ready, then over to you. Okay, thank you very much Martina for the introduction. I also want to say thank you to you and your colleagues at ONS for all the hard work you do to produce this really useful dataset. You know, we've used a huge amount in our team. So this work I'm going to present today is joint work with Mark Bryan, Nigel Rice, Jenny Roberts and Christina Sechel and this is funded by the Health Foundation and hopefully you'll have a chance to ask questions and to interact with me later you can see my Twitter address down there so please do track me down on Twitter. So firstly a summary of our research. So as we all know in the start of 2020 we had the COVID pandemic started and it was a huge recession, probably the biggest recession we've ever had in living memory. But unlike many recessions it didn't involve a huge fall in the employment rate because of the coronavirus job retention scheme which was introduced in order to protect people's jobs but what we did see is a huge number of people being temporarily away from work for example being on furlough. And what we do find is that people with mental health or physical health disabilities have been more likely to be away from work or working reduced hours during the pandemic. And what we do in this paper is to decompose that into what part of this gap can be explained by different characteristics. For example disabled people being overrepresented in part-time work or occupations that can't be done at home. But we also find that an unexplained component remains which suggested that disabled people have either been treated or are behaving differently even though they are in the same types of jobs. So the background to this, well disabled people in the UK are of working age are much less likely to be employed than non-disabled people. The gap is particularly big for people with a mental health disability but it's also very big for people with a physical health disability. And reducing this gap is a major government policy. The aim is to increase disabled employment by one million between 2017 and 2027. And this was the subject of a work in pensions committee inquiry about a year ago which our team contributed to. Now this employment gap has been narrowing since 2013 but how will this trend be affected by Covid? Now clearly there's been a lot of research on the labour market effects of the pandemic and the fact that it has affected different groups unequally and much of the evidence has been based on things like ethnicity, age and gender. There's significantly less been written about how disabled people have been affected although Emerson et al. 2021 and Jones 2022 are exceptions. And as I said before the job retention scheme has cushioned the blow of the pandemic but the fact that some people have been more affected, have had their jobs restricted may have possible long-term effects. So let's just look briefly at the theory. Why is it that disabled people might have worse labour market outcomes than non-disabled people? Well part of this might be to do with discrimination. So employers may discriminate against disabled people either directly simply due to prejudice or discrimination can also happen indirectly. For example employers may incorrectly assume that disabled people have lower productivity or are more likely to leave the job and this can lead to unequal outcomes. However part of the story may not be about discrimination within jobs but just the fact that disabled people are concentrated in particular sectors or particular types of job. And so we've classified this as segregation and segmentation. So segregation means that there's unequal distribution of disabled workers across occupations and or industries. Now this is a very common explanation for explaining something like the gender pay gap whereby within a particular job maybe men and women are being paid the same but because women tend to be more concentrated in particular types of occupation or particular types of industries where pay might be lower this is what explains the gender pay gap rather than directly discrimination in many cases. And we propose that and we find in this paper the same thing happens for disabled workers compared to non-disabled workers. And we also look at segmentation which is very similar to segregation but this is about the unequal distribution of disabled workers across different contractual arrangements. Again when we look at gender we know that women are much, much more likely than men to work part-time and part-time workers tend to be more vulnerable. For example when there is a recession there often the workers that get laid off at first and therefore this can explain why women might be more vulnerable in the labour market than men. And again we use this same theory and the same ideas to explain the difference in outcomes between disabled and non-disabled people. So our hypothesis that we're testing in this paper is that we suggest that COVID-19 has widened the disability gap not in employment as such but in these variables away from work and reduced hours which I'll talk about in the next slide. And these gaps should be partly explained by measurable factors relating to segmentation and segregation. That includes industry occupation, workplace size, public sector affiliation and part-time status. But part of the gap will remain unexplained and this could be caused by discrimination which I talked about before but it might not be that. It could be that it actually reflects worker preferences. So if disabled people are more likely to be going on furlough than non-disabled people this could be to do with their preferences. Maybe they're clinically vulnerable or more likely to be clinically vulnerable and actually are wanting to stay at home or go on furlough or reduce their hours. Or it might be that their employer is pushing them towards that and maybe are pushing disabled employees into furlough more than non-disabled. So let's now look at the data. So we use the quarterly LFS. And in this decomposition analysis which I'm going to present we focus on quarter two of 2020 and compare that with quarter two of 2019. And we focus on three labour market outcomes. Whether in employment and whether temporarily away from work and also whether working reduced hours due to being laid off short time or work interrupted by economic and other causes. This question focuses on people who say that they've done the actual hours they did in the reference week was lower than their usual hours and this was the reason that they give. And we find that this response has been used quite a lot since the pandemic began. And we also exploit the fact that the LFS has a really nice definition of disability based on the Equality Act 2010 which is whether they had a health any health condition or illness lasting 12 months or more which reduces ability to carry out day-to-day activities. So disability status is a derived variable in LFS and that is our primary variable that we use, our explanatory variable that we use. But we also look at the particular condition that they report and we use that to split this into mental health and physical health. Our method is the Oaxaca-Blinda decomposition method. I'm not going to go through this slide in any detail but just to give you the intuition. So if we're interested in finding out the difference between the labour market outcome Y of disabled people versus non-disabled people we find that this can be broken down into what we call characteristics. So these X's are the particular observable characteristics of the individual and let's say that X stands for education. We know that the more education you get the higher returns, the better labour market outcomes you have. So if, for example, disabled people are less likely to have a degree that means that because the returns to a degree are higher than the returns to lower level qualifications this would explain why there might be a gap in these outcomes for disabled people because they're less likely to have the kind of characteristics which predict good outcomes. And then we have this other component which is coefficients. So again, thinking about education, we might see that on average disabled people with a degree actually have worse outcomes than non-disabled people with a degree. So the returns to education are lower for disabled people which this could be evidence of discrimination. So what we do is we break it down to the characteristics, sort of the explained part and the coefficients which is the unexplained part. And these are the different variables we use. So the X variables, we have education and experience, we have 19 categories of industry, 9 categories of occupation. That's your segregation industry and occupation. And the segmentation side is size of workplace, self-employed status, where the public or private sector and where the part-time or full-time work. And we also look at region and demographic controls. So now let's have a look at the results. First of all, some descriptive results. So here we see that the disability gap in employment is very big, as I saw before, particularly with regards mental health, which is this pink line at the bottom. But we see that after the pandemic, which is shown by this vertical dotted line, there was actually very little change in the employment rates of men and women and also in the gap. That's because the job retention scheme protected people's employment, and so very few people were actually laid off. But when we look at these other outcomes away from work, we see this massive spike in 2020 quarter 2 in people who were temporarily away from work. And we see that the gap actually increased between disabled people, which is the top lines here and the black line, which is our comparison group non-disabled people. And that was the same for men and women. And similarly with reduced hours, we see that from almost a baseline of almost zero before the pandemic, a number of people on reduced hours increased more for disabled people than black line, which is non-disabled people. And we can also look at how people are concentrated in different occupations in industries. And we see that non-disabled people tend to be more heavily concentrated in professional occupations. And disabled people, particularly those with mental health disabilities, are more concentrated in elementary occupations and for women, particularly in caring leisure and other service occupations. And we see a similar picture by industry. So, and then again, it's particularly mental health disabled people who are affected. They are more likely to be employed in wholesale and retail and more likely to be employed in accommodation and food services. So already we can see this kind of segregation going on, these kind of jobs, which people couldn't do at home, which resulted in people being put on furlough. Disabled people were already more concentrated in these jobs and say that we're going to be more vulnerable to lockdown. And we can also then have a look at the particular gaps between disabled and non-disabled people. So, while in the top graph here we see the gap in employment, while that is particularly big for those with disabled, mental health disabled, there's not really any change between 2019 and 2020, where we do see the change is in away from work, where the gap actually increases both for mental health and for physical health. And we also see a gap of about 5 percentage points in the number of people, the number of employed people who are working reduced hours. This is in quarter two of 2020. So, what we do then is, sorry, this is now the female gaps and this is similar, so it's a similar picture, except that we see the gap in reduced hours is a little bit lower for women than it is for men. But in general terms, the picture is the same. The gaps have increased as a result of the pandemic. So, what we then do is decompose this into characteristics and coefficients. And the highlight of this graph is that this orange bar, which is the coefficients, is the main part. So, this is the unexplained part, the part which might be to do with being treated or behaving differently because you're disabled. What we do see is that the blue part, particularly on the mental health side, which is the characteristics, actually becomes more important after the pandemic. It becomes more important in 2020 than 2019, which does show that segregation and segmentation have been playing a role more so because of the pandemic and because that has affected different types of jobs in different ways. What we also do, as well as just looking at this overall decomposition, we drill down and look at the particular observable characteristics which explain these gaps. So, there wasn't that much. So, what we show here are the statistically significant gaps, the characteristics which are statistically significant in terms of explaining the gaps. And we see that there's much more here in the away from work gap. There are many more explanations here. So, we see that segmentation is very important. This is particularly related to part-time work. So, for example, about 8% of non-disabled men who are employed work part-time, whereas 22% of men with a mental health disability work part-time. And so, because part-time workers were much more likely to be away from work during the first lockdown, this explains a lot of the gap. In fact, for all of the gaps, men and women, and both mental health and physical health. Occupation is also very important. And here, we find that disabled people are more concentrated in elementary occupations or caring and leisure and other service occupations, particularly females, and are less concentrated in managerial and professional occupations. So, the kind of occupations which people had to go on furlough because they couldn't work from home, because their workplace had been shut down. This is where disabled people were concentrated. And this is explaining significantly this away from work gap. And we also see the same with education and experience. Again, this is about having a degree. People with degrees or higher qualifications tend to be able to work from home and were less affected by the pandemic. And disabled people were less concentrated in those areas. And we do see some similar effects explaining the reduced hours gap as well in 2020. But still, I would emphasize that coefficients, i.e. the unexplained component, still explains most of the gap, which does show that even within jobs, within sectors, disabled people were being treated differently and did have different outcomes. So, I'll finally move on to my conclusions and then we can move on to questions. So, while employment has generally been protected, we see that the pandemic has exposed new inequalities in the outcomes of disabled workers, namely, propensity to be away from work or on reduced hours. So, a relatively small proportion of this is due to characteristics that can be observed. But this component did increase in 2020, so the pandemic has exacerbated the segmentation and segregation of disabled workers. And this is particularly due to disabled people being more likely to be in part-time jobs and having uneven distribution across occupations. There's less of an evidence of a segregation effect, although men with mental health disabilities were more likely to work in accommodation and food services, and this did affect their likelihood of being away from work. So, most of the gap remains to do with coefficients, which means that this could be due to employer attitudes. So, for example, risk-averse employers may be more likely to want to temporarily lay off or furlough disabled people, or it could be to do with worker preferences, disabled people wanted to go on furlough to protect their health. So, what does this mean long-term? Well, we're not quite sure yet, but in the post-JRS world, this could involve structural unemployment as redundant jobs are no longer supported, and this may disproportionately affect the employment prospects of disabled people. So, the focus should be on ensuring disabled people have the training and skills needed in the restructured post-COVID, post-Brexit economy. Okay, thank you. Thank you very much, Andrew. Really interesting piece of work. We'll move on to our next presenter, Magdalene. Magdalene Okulou from the University of Bath. She'll be speaking about modelling the differences, the different impacts of COVID-19 in the UK labour market. Magdalene is a lecturer and researcher in macroeconomics, and she completed her PhD and began lecturing in the University of Bath in 2021. Her research interests include DSGE and labour market search modules, and the UK gig economy. Magdalene, if you're ready, then I'll hand over to you. Yes, thank you. Can I just confirm that you can see my screen? Yes, we can. Thank you. Good afternoon, everyone. Thank you for joining me. And thank you to the organisers for the opportunity to present my research. I will be discussing today the paper on the different impacts of the COVID-19 pandemic on the UK labour market. It's a paper I have co-authored with Chris Martin, and we have recently revised and resubmitted it to the Oxford bulletin for economics and statistics. So in the next 20 minutes, I will talk you through our research motivation. I will spend most of the time on discussing how we analyse the data. I will run through the model and conclude with our baseline results and some scenarios. So why did we choose to write this paper? The COVID-19 pandemic had a massive impact on the UK labour market, but in particular, before the pandemic even, there has been evidence in the data that graduates were more likely to earn higher wages and be in more stable jobs than non-graduates. And the data shows a lot of job-to-job movement. Workers are more likely to move from job-to-job than from unemployment into employment. So putting all these points together, we try to see if we could use a relatively simple model, macroeconomic model to describe the extreme turbulence induced by the pandemic. We also try to see if we could analyse the differing impacts of the pandemic on different types of workers in the UK. And also we try to measure the impact of the job retention scheme. So our results closely match the results for output and employment throughout the pandemic. Analysis also shows that non-graduates were worse during the pandemic than graduates. And our results suggest that the JRS job retention scheme saved or prevented the loss of between four to five million of the furlough jobs from being lost. And most of the saved jobs were low-wage jobs, mostly held by non-graduates. So how does our paper differ from the rest in the literature? What we did was to construct a DSG model with labour market search frictions. Our model has four distinct labour markets for non-graduates in high and low SSEC occupations. I'll talk more about high and low SSEC occupations in the coming slides. We also model job-to-job movements and our model tracks about distinct worker transitions. Most importantly, we have a richer combination of shocks to describe the pandemic. We model aggregate demand and supply shocks, but we also have job-specific shocks to capture the evidence that workers in low SSEC employment were more likely to be furloughed and furloughed for a longer period than those in high SSEC employment. And we also model shocks to job destruction induced by the pandemic. Now moving on to the data. We use the five-quarter longitudinal LFS data before the pandemic. We used data from the end of 2018 to the end of 2019. And for the pandemic period, we used the end of 2019 to the end of 2020. For graduates, we split workers into graduates and non-graduates. Graduates are those who have at least the first degree and non-graduates are all the others with less than the first degree. For high productivity jobs, we classify those as those in SSEC groups one to three. And low productivity jobs, we classify those as the rest, that's groups four to nine. Now for the period before the pandemic, we did a small analysis on the data set. And this pie chart here shows us that non-graduates are highly likely to be out of work than non-graduates. And the pie chart also shows that there's a substantial number of non-graduates who have high SSEC jobs, but there is an even greater proportion of graduates who have high SSEC jobs. And in particular, we observe that there's a greater proportion of non-graduates in low SSEC jobs than the number of graduates in low SSEC jobs. Moving on to transitions before the pandemic, our table here, we tracked movement, the movement of workers from employment into non-employment and the reverse. What we noticed in the data is that there was a lot of movement between unemployment and inactivity and the reverse, and a lot of movement from inactivity and unemployment into employment and the reverse. So what we did was to bunch the inactive and the unemployed into one group, which we call the non-employed. So we tracked movements of workers from different states of employment through the five quarters of the pre-pandemic period and took an average, which is what we displayed in this table. From the table, we see that low SSEC jobs are more likely to break down for both graduates and non-graduates. We also can see from the table that non-graduates are more likely to remain in low SSEC jobs than graduates. Also, we see that graduates are more likely to remain in high SSEC employment than non-graduates. And graduates and non-graduates are more likely to move from job to job than they are to move from non-employment into employment. And finally, we find that non-graduates are more likely to remain out of work than graduates. Now to the pandemic period, we see from our table that employment seems fairly stable across all four quarters of the pandemic. But I think we can see a clear picture of what happened during the pandemic in this next chart here, where we track the changes in the levels of employment of different categories from quarters to quarter in millions. And here we see that the non-graduates in low SSEC employment were the first and hardest hits by the pandemic. And the impact of the pandemic comes on for other types of workers later on in the year. Now, if you caveat about the LFS data we used, there are suggestions that the LFS may have understood the extent of job loss during the pandemic for a couple of reasons. One of which is that some workers may have changed their employment status from self-employed to employee in order to access support under the government scheme. That factor does not affect our analysis because we analyze both the self-employed and the employees. Also, there were workers who were earning no wage and were not furloughed but still reported themselves as employed. For example, business owners who could not access support under a government scheme or zero contract workers, for example, such workers would there was a bit of a gray area between what was called being employed and being unemployed or non-employed during the pandemic. And still on furloughing the LFS data we used did not collect information on the workers that were furloughed. There was the understanding, society survey data which collected furloughed data but that data set was we couldn't get all the categories or the characteristics of workers that we wanted for the people with that data set. So we stuck with the 5Q longitudinal LFS data. However, in this data set we used we here in this chart we see that from measuring the professional percentage of workers who reported themselves as employed but were doing new hours we see a spike from the beginning of 2020 indicating some workers being furloughed and the highest spike we see is of non-graduates in low SOC employment. It gives us a feel of the extent of furloughing and the extent of workers who were just doing zero hours while reporting being employed. Now moving on to the model as I mentioned earlier workers can be either employed or non-employed and the non-employed are some of the unemployed and the inactive. We assume that all workers search and all the categories of workers non-employed either being non-employed in high productivity or low productivity employment. These categories are we have adopted the numbers from all the ones we have seen in the table earlier. Also the red arrows indicates the transition of workers from one state into the other and highlights the transitions of workers we just saw in the table earlier. We also assume that there's one type of firm that offers four types of jobs for graduates and non-graduates in high and low productivity types of employment. We assume that productivity varies from job position to job position and that the firm bargains with each type of worker which gives us four distinct wage equations. We also assume that job destruction is exogenous. I will speak about that at the end of the presentation. So how we calibrated the pandemic shocks. We assume that the pandemic would have induced a the pandemic did induce a fall in aggregate demand but we assume that the job retention scheme reduced that fall in aggregate demand. We assume that the pandemic would have brought on a fall in aggregate supply from social distancing from working from home and so forth. But we assume that the job retention scheme would have worsened that negative impact because of furloughing. We also assume that because of furloughing and the fact that some workers especially those in low productivity employment were more likely to be furloughed and furloughed for a longer period we reflect that by an additional fall in productivity for low SOC jobs. We also assume that the pandemic would have induced job destruction or increased job destruction but the job retention scheme would have reduced that negative or that impact on job destruction. And finally we assume that the pandemic would have induced a fall in wages but the wage subsidy under the job retention scheme would have mitigated that fall in wages. So all of this we have just discussed what we have outlined in the baseline column here. Now how to measure the impact of the job retention scheme. What we did was to in order to measure the impact of job retention scheme we tried to imagine a scenario where the job retention scheme did not exist and compare that to the baseline. So we actually tried four different scenarios but I'm for the sake of time just going to present two scenarios. In scenario one we assume that in the absence of job retention scheme productivity would have fallen but just by a very small amount compared to the baseline there would be no wage subsidy and we assume that the falling aggregate demand would have been similar to the baseline and job destruction of course would have been higher. In the second scenario we tried to imagine the reverse now we imagine that in the absence of the job retention scheme productivity would have fallen by a lot but by an amount similar to the baseline as before there would be no wage subsidy and we assume that in the absence of the job retention scheme the impact on aggregate demand would have been a lot double the the impact in the baseline and we also assume that job destruction is higher. Now quickly moving on to the results the results the graphs in the straight line show the results of our simulation whereas the graphs in the dotted the dotted lines indicate the actual so our results closely match the results for outputs during the pandemic and recovery for employment our results fairly match the actual but our results also show the relatively small fall in employment for both graduates and non-graduates throughout the pandemic and the results for wages as well shows the fall in real wages at the peak of the pandemic and the spike in wages during the recovery period due to composition effects. Now to analyzing the impact of the job retention scheme in the graphs shown here the straight lines are the results of our simulation the dashed lines are the results of scenario one and the dashed and cross lines are the results of scenario two so what we see here is that the results for scenarios one and two are quite different from each other although the scenarios are a bit less obvious of themselves but the results for graduates and non-graduates employment are a bit more similar, more consistent which gives us a bit more confidence about the outcome for employment than for output and what we see here is that how we calculated things is the difference between the scenario results and the baseline results gives us the number of jobs that might have been saved by the job retention scheme and this translates to between four and five million jobs most of which would have been those held by non-graduates so finally our results show that a relatively simple DSG model can give some insight into what happened during the pandemic and can be useful for policy analysis however we made a series of strong assumptions for example we assumed that the economy was in steady state before the pandemic this assumption was convenient for us to make because what we were looking at is the impact of the pandemic on the labour market but what we could do better in future research might be to analyse the labour market for a longer period before the pandemic see the decade before the pandemic and compare that to the data during the pandemic also we assumed that the economy goes back to or the economy will go back to steady state and there's no permanent impact of the pandemic on the labour market which at the moment we still do not cannot be sure of also our results for inflation don't match the actual inflation this is because we assumed that inflation is a fixed markup on the marginal cost and because we modelled the wage subsidy due to the job retention scheme the marginal cost was relatively flat and so inflation was relatively flat as well so going forward it would be interesting to analyse the impact of the pandemic on other types of workers for example the gig economy gig workers, zero hour contract workers and so forth and it would also be interesting to analyse the impact of the pandemic from the from the aspects of the firms and how firms were adapted through the pandemic in particular we assumed that job destruction was exogenous and then modelled the shock representing the impact of the pandemic we could model the decision of the firms to destroy a job furlough worker or not furlough worker for example so those are things that we can consider in future work so I will leave it here for now and take your questions okay thank you so let's move on to our third presenter in this session Francesca Foliano she works at the UCL social research institute and she is a research fellow her main research interests are in economics of education, labour economics family economics and international trade so Francesca if you are ready I can handle to you I think you still knew that I can hear you so I okay can you see my slide yes we can thank you okay so thank you very much for staying for this presentation and thank you very much to the organisers for allowing me to present these work this is a project that has been going on for a few years now and with my colleague Rebecca Riley from King's College this paper is titled global competition UK labour market adjustment at the Brexit vote and this part of a bigger project funded by the Nuffield Foundation so in this research our aim was trying to understand how the UK labour market adjusted to the sharp rise in imports from low wage countries in particular we focused on imports from China and the Eastern European countries that joined the EU in 2004 and what we look at is how these imports affected mainly the manufacturing sector but then more widely other labour other labour outcomes in local labour markets and also workers' mobility across areas and voting patterns we focus on the populist vote in particular vote for UKIP and the Brexit vote in 2016 so before going on I just wanted to present the usual disclaimer about the data and I take advantage of this to say thanks to all of the UNS for the amazing work they do I must say I mean these data are great I use the labour for survey and ask data for this work and they're great thank you thanks a lot ok so motivations why do we care about these there are lots of papers already in the economics literature that look at the effect of trade exposure increasing import competition from China in particular on local labour markets there are some papers that look at different outcomes in the US papers that look at different countries in the EU so why do we care about another paper what is interesting is that the way labour markets in different countries have adjusted to change in international trade differed according to institutional settings and the economic structure of the country and understanding the mechanisms through which in the UK local labour markets adjust to structural changes is important in particular because Brexit has already started to bring such some of these structural changes already but also because the green transition is likely to bring more and now after two years of pandemic we might face other challenges so understanding how the local labour markets adjust to employment shocks is important because it can help designing more targeted policies that can contrast the inevitable inequalities that comes with structural changes in the economy ok so why do we care about China why do we focus on China and the 8 countries it is the Eastern European countries that joined the EU in 2004 so what happened was that China, both China and Eastern European countries China from the 80s Eastern European countries from the 90s experienced an incredible increase in labour productivity and this made them ready to respond to international trade international demand for goods in a very fast way here I present just a descriptive trend in trade the UK had with China and the 8 countries and here you can see that the import share from China increased from 95 to 2015 by a factor of 5 and the imports share from 8 countries increased by a factor of 4 at the same time the exports from the UK to China increased as well by a factor of 4 whereas the exports from the UK to 8 countries didn't experience a particular change but clearly this increase in imports from these countries, these low-age countries happened in a particular moment for the UK, in a moment in which the manufacturing sector experienced a shrunk, in particular if we look at the same period, 95 to 2015 we see that the share of jobs by industry decreased by from 16% to 7% like a great decrease that was not experienced by other industries in the UK over the same period it is also important to realise that this shrinking of the manufacturing sector happened in a it was part of a long-term decrease in manufacturing jobs partly driven by technological changes changing tastes and also the explosion of the service sector that was happening at the same time together we of course import competition from other countries and so the aim of our paper is mostly to disentangle how much of the change in manufacturing jobs is driven by import competition from China and A8 and then look at the mechanism which local labour markets adjusted to these shock so these are the research questions in our paper did China and A8 shock affect the UK manufacturing sector in the long-term how did local labour markets adjust over time and did increase in trade exposure of local labour market affect the rise in the populist vote and the vote to leave the EU so these are the main results that will present the questions and the relative results that will present today so what we do in this paper as I said we focus on the period 2000-2015 during this period we know that China joins WTO and the A8 joined the EU but as I said their productivity growth had started way before then the local labour markets their unit of analysis are travel to work areas this is how we define our local labour markets we define a measure of change in local import penetration that follows author Dorn and Hanson who have a seminal paper on the effect of import exposure on US manufacturing sector and we estimate the effect of import exposure on local labour market outcomes for the long run by accounting for possible endogeneity of these trade shocks because we use an instrumental variable estimator and we investigate the influence of trade shocks on the populist vote very quickly a preview of the results what we find is that an increase in import competition between 2000-2015 from country with relatively low pay resulted in a decline in total manufacturing jobs workers moved out of manufacturing into low-skill low-paid non-manufacturing job and this change was associated with a decrease in mean today I will present median actually but we have the same result for mean as well, mean and median weekly earnings in non-manufacturing sectors particularly in low-skilled occupations we also find an increased skill polarisation across different areas of the UK that is basically we observe that skilled workers move away from areas more affected by import competition to other areas less affected in addition we observe that actually import competition has very little effect on the populist vote if any and actually this is more driven by long-standing economic characteristics of the areas that are more voted for and live during the Brexit referendum and possibly also driven by the movement of more skilled workers away from these areas okay how do we contribute we contribute to the literature by presenting a comprehensive study of the effect of increased trade exposure on local labor market outcomes in the UK and we explore in particular the effect of import penetration on internal migration which hasn't been done before and we link these changes to recent election results I'll skip this but there is a wide literature that looks at trade competition on local labor markets firms and election results and we contribute to these presenting some results for the UK so methodology we use as I said the methodology pioneered by author Dornan Hanson and basically this methodology is based on two questions they asked where in the country did the UK produce before the trade shocks the goods that now are imported from China and the A8 and what has happened in those parts of the country relative to other parts of the country since the steep increase in trade exposure so it is interesting to focus on China and the A8 as I said before because the increase in imports from these countries was mostly driven by a supply shock the fact that these countries were having political changes that made them more productive definitely there was also a domestic component that we try to let's say address with an instrumental variable estimation so how do we define for each local labor market a measure of import exposure per worker so basically in order to create this measure the idea behind this measure the idea that author Dornan Hanson had was that the exposure of local labor markets to trade penetration differs according to their initial industry employment specialization so in this case the change in import competition for the local labor market I is given by the sum overall industries J of the total change in imports from change from China and A8 countries between 2000 and 2015 weighted by the local share of employment by industry J and then this measure is then divided by the number of total jobs in the local labor market so this gives a monetary measure per worker of how much the area is exposed to import competition from these low wage countries we consider also alternative measures in particular measures that take into account competition in international markets as well because of course the UK not only competes with China and A8 countries in the domestic market also international markets and also a net measure that takes into account experts from the UK to these countries but the results are broadly similar so I will focus on the results for import competition only then once we have this measure we estimate the following equations so on the left hand side we have a change in local labor market outcomes and in case of votes we have a change in UKIP votes in UKIP votes for the European elections and on the right hand side we have the measure of import exposure plus a comprehensive set of controls at the local labor market level and the list of controls are here and mostly we include everything that can describe the initial conditions and the changes in the local labor markets during this period 2015 so as I said our beta the parameter of interest which is the impact of import competition, the change in import competition on local labor market outcomes relies on the fact that this change is exogenous so it's only due to supply shock however we know that this might not be the case and there might be reasons for internal reasons for which also the demand for imported and domestic goods changes so we take into account this with an instrumental variable estimation where we instrument the measure represented before with a similar measure where the total change in imports is now given by the change in imports from other European countries in particular we focus on the U15 minus the UK and the share for the share of employment by industry for each local labor market is instead measured in 2000 is measured for 1995 we use several data sets for this analysis but mostly we use the labor force survey and the data and the labor force survey was used to create all of the local labor market outcomes and the measures so all of the shares that are part of these measure of import exposure that I discussed so far okay so the final data set basically we have a final data set that has 232 observations one for each travel to work area and measure changes in trade exposure and labor market variables between 2000 and 2015 okay so just to give an idea of this measure of import competition so basically the mean of this measure across the UK is 1433 pounds so this is the import exposure per worker the mean import exposure per worker in the UK between the periods 2000 and 2015 here we can see that considering also international imports increases sorry competition in international markets increases this measure at the same time when we take into account the exports this measure is decreased but is broadly similar and this gives an idea of how this measure differs across the UK okay so let's I'm going to present now the main results of this analysis so a change in import competition determines as we can expect a change in manufacturing jobs per head and manufacturing jobs so increase in import competition means that manufacturing jobs per head and manufacturing jobs decrease in the areas more exposed to import competition we don't see any effect on non-manufacturing jobs per head and we see a total effect driven by the manufacturing jobs so areas that are affected by more trade competition compared to others see a decrease in the total number of jobs driven by a decrease in jobs in manufacturing for what it concerns the median weekly earnings we observe that these are quite interesting so we observe that in the manufacturing sector the earnings actually go up so areas that are more exposed to trade competition to import competition see an increase in the median weekly earnings and this is a compositional effect due to the fact that low skilled workers that were laid off over time were mostly low skilled workers but then when we focus on other sectors in particular low skilled non-manufacturing we see that there is a decrease in median weekly earnings in the sectors too and this is clearly driven by the fact that low skilled workers who were laid off by the manufacturing sector were then absorbed by low skilled non-manufacturing sectors determining lowering of the weekly earnings and we see this effect more general for all median weekly earnings in areas more affected by import competition so there is a general depression of wages in these areas compared to others then we see that actually in the long term there is no effect on an unemployment rate and in our analysis we look also at the medium term there is an effect on an unemployment rate unemployment rate goes up but then when we look at the long term that is 15 years the unemployment rate is not affected and this is again a result of the fact that the low skilled workers were they were absorbed by a low skilled non-manufacturing sector then we look at the population changes so we are interested in seeing whether these trade shocks can affect on the population in areas more affected by them so we see that the total population seems to decrease however this result is not statistically significant but what we observe is that there is a clear decrease in the population with degree level so it looks like that people the skilled workers some skilled workers once these areas are affected by a trade competition leave the areas and we observe this in these results and also we see that the shadow working age population with degrees decreases over time that these areas more affected by important competition see a decrease in the shadow working age population with degrees relative to other areas we perform also an analysis using hash data at individual level basically I'm not going to go into the details but what we look at we use longitude in a hash data to see how whether individuals who we can observe both in 2000 and 2015 are more likely to move away from areas that were more affected by a trade competition and we see that actually this is the case that individuals who are in areas more affected by a trade competition were more likely to move away and this is particularly true for high skilled workers and workers with high ability to earn I'm going to skip these otherwise all the time but very quickly I want to talk about the UKIP votes the results for the UKIP votes so when we look at the changes in UKIP votes in European Parliamentary elections between 2004 and 2014 on the import competition measure we observe that there is an association but this association disappears once we control for a comprehensive set of characteristics of the local labor market and this seems to suggest that actually was not a trade competition but rather longstanding economic issues that have driven the populist vote in areas more affected by import competition and very similar results we obtain very similar results for the EU referendum in 2016 once we control for a comprehensive set of results there is no association and given the relationship between the share of population with degree and the share of live votes it is likely that what we can say possibly is that the trade competition affected the populist vote more by moving skilled workers away from certain areas workers that were probably more likely not to vote for UKIP or live in the Brexit referendum and so I'll go very quickly the conclusion so in this paper we studied the effect of trade shocks on the UK labor market we observed the longer trend consequence for unemployment that were mitigated as lower skilled workers moved to non-manufacturing sectors and took lower pay higher skilled workers moved away from more exposed areas and this last result partly might explain the correlation between import exposure and the live vote and the increase in the share of live votes for UKIP thank you very much