key: cord-0065008-twbo6p1i authors: Blit, Joel title: Automation and Reallocation: Will COVID-19 Usher in the Future of Work? date: 2020-08-01 journal: nan DOI: 10.3138/cpp.2020-065 sha: 0748509e4227dbb38d225d5a7d5d2d99c8cb6735 doc_id: 65008 cord_uid: twbo6p1i Recent evidence for the United States suggests that recessions play a crucial role in promoting automation and the reallocation of productive resources, which in turn increase aggregate productivity and lead to a higher standard of living. I present evidence suggesting that the same is true in Canada. In particular, since the beginning of the information and communications technology revolution, all of the decline in routine job employment occurred during the subsequent three recessions. A similar dynamic is likely to operate during the COVID crisis, and in fact is likely to be more pronounced due to the scale of the recession and the health-related incentives to automate. By constructing industry-level measures of worker exposure to COVID and the fraction of routine employment, I show that the retail, construction, manufacturing, wholesale, and transportation industries are likely to experience the biggest transformations. In these industries, government attempts to maintain the status quo will only delay the process of restructuring. Instead, policies should embrace change and support workers through the transition. COVID-19 has engendered the biggest human and economic catastrophe in recent history. It has resulted in over half a million deaths worldwide and shut down entire economies. At its peak in April, more than 5.5 million Canadians either had lost their jobs or were working less than half of their regular hours. Worse, it has disproportionately impacted our most fi nancially vulnerable: low wage earners, part-time employees, the solo selfemployed, youth, and recent immigrants. It is thus diffi cult to entertain the idea that the crisis may come with silver linings. One of these is the increased solidarity that has developed from the realization that we are all in this together. Another is the acceleration of automation and reallocation that will increase productivity and the living standard of Canadians. It is the latter legacy, and how best to implement policies that address the realities of the crisis but do not stifl e economic transformation, that this article addresses. The notion that one should "never waste a good crisis" is particularly apt today. Economic recessions raise the incentives of fi rms to increase effi ciency through automation and reorganization. The shock also results in a reallocation of productive resources from less productive between internal innovations that improve a fi rm's existing product lines and external innovations that allow fi rms to capture markets from others ( Akcigit and Kerr 2018 ) . Internal innovations increase aggregate productivity directly by making fi rms more productive. External innovations increase aggregate productivity through the reallocation of resources from less to more productive fi rms, whether within the same industry or not. The extent to which aggregate productivity growth is driven by within-fi rm productivity increases versus between-fi rm reallocation has been the subject of debate for decades (see Foster, Haltiwanger, and Krisan 2001 for an overview of the early literature). Their relative importance may depend on the sector and the historical period in question. For example, Foster, Haltiwanger, and Krizan (2006 ) fi nd that productivity increases in the 1990s retail industry were primarily driven by reallocation between fi rms, and in particular were due to entering large national retail chains. Consistent with this, Akcigit and Kerr's (2018 ) model fi nds that the contribution of external innovation to aggregate productivity growth is between two and three times larger than that of internal innovation. On the other hand, a broader study of private businesses over a thirty-year period by Garcia-Macia, Hsieh, and Klenow (2019 ) concludes that within-fi rm innovations are relatively more important. For Canada, Baldwin and Gu (2011 ) fi nd that in manufacturing within-fi rm productivity growth is relatively more important than between-fi rm, while in retail the opposite is true. These fi ndings are representative of the wider literature in that both withinfi rm innovation and between-fi rm reallocation appear to be important in raising aggregate productivity. Thus, in thinking about the impacts of COVID-19 and possible policy responses, both must be considered. Schumpeter's insights into the dynamic nature of a capitalistic economy also extended to the important role of recessions in bringing about productivity-enhancing transformational change. In his book chapter "Depressions: Can We Learn from Past Experience?" ( Schumpeter 1934 ( Schumpeter [1989 , 16), he writes that "depressions are not simply evils, which we might attempt to suppress, but . . . something which has to be done, namely, adjustment to . . . change." This view, that recessions are required for restructuring the economy, was largely forgotten over the decades that followed. Davis and Haltiwanger (1992 ) once again sparked interest in the impact of recessions on aggregate productivity growth by showing that job reallocation (and especially job loss) disproportionately occurs during recessions, and tends not to be industrywide, but rather focused in specifi c plants. To explain these results, models need to explicitly allow heterogeneity across fi rms in a given industry, and this heterogeneity needs to interact with the business cycle. One class of models that can generate such results has fi rms with different abilities to (or maladapted) fi rms to more productive ones, increasing aggregate productivity. A number of recent articles have found evidence for the United States that is consistent with these mechanisms. Hershbein and Kahn (2018 ) show that automation and demand for skills rose most in regions that were hardest hit by the Great Recession. Jaimovich and Siu (2020 ) demonstrate that since the start of the information and communications technology (ICT) revolution, the downward trend in routine employment has been driven almost entirely by permanent losses during recessions. This article presents similar evidence for Canada. In particular, I fi nd that all of the routine job losses in Canada have occurred over the last three recessions, and that these jobs do not return. Non-routine employment, on the other hand, is stable during recessions. This suggests that recessions play a crucial role in fostering automation and reallocation in the Canadian economy. This article also discusses similarities and differences between our current self-induced recession and past ones. This crisis will surely result in a bigger economic transformation, not just because of its scale, but also because fi rms have added incentives to automate to protect the health of workers and mitigate risks to their operations. These COVID-specifi c incentives to automate differ across industries because workers in different industries face different levels of exposure. In addition, the feasibility of automation differs across industries, as some have a higher prevalence of routine tasks than others. By quantifying workers' level of exposure and the fraction of jobs that are routine, I show that retail trade is the industry that is likely to experience the biggest transformation, followed by construction, manufacturing, wholesale trade, and transportation. Policies that attempt to maintain the status quo would thus be particularly damaging in these industries. The article proceeds as follows. Recessions and Automation presents a literature review on the impact of recessions on aggregate productivity growth. Recessions and Automation in Canada presents evidence for Canada suggesting that automation and reallocation occur during recessions. Is the COVID Recession Different discusses the extent to which this recession is likely to have similar (or bigger) transformational impacts as recent recessions, and Industry Automation Incentives and Opportunities shows how this might differ across Canadian industries. The last sections are Policy Considerations and a Conclusion . Since at least Schumpeter's (1942 ) writings on the gale of creative destruction, we have known that our economy is a dynamic entity where process and product innovation drive the incessant creation and destruction of products, jobs, and fi rms. This innovation, whether it be process innovation that lowers input costs or product innovation that results in higher-value-added outputs, propels aggregate productivity growth. We can also distinguish De la Rica, and Lacuesta (2014 ) document a similar acceleration in job polarization due to the disappearance of routine jobs during the Great Recession in Spain. Zhang (2019 ) shows that during economic downturns, fi rms that employ more routine labour invest more in machines and decrease routine employment. Taken together, the empirical evidence overwhelmingly points to recessions as important periods of accelerating automation and therefore of increasing productivity. Not only do we observe a relative increase in investments focused on automation, but also we observe job upskilling and a permanent decline in routine jobs. With the exception of one article focused on Spain, all of the evidence discussed so far is for the United States. To the best of my knowledge, no similar Canadian studies exist. In this section, I follow a methodology similar to that of JS and show evidence that in Canada, too, recessions accelerate job polarization. I obtain data on monthly employment by 2-digit National Occupational Classifi cation (NOC) from the Labour Force Survey (LFS) for the period from January 1987 to May 2020. 1 I adjust the data for seasonality and apply a low-pass fi lter to remove any fl uctuations with frequency higher than the inverse of 18 months. 2 This produces employment curves that retain their shape but are smoothed. For each occupation, I determine whether the associated tasks are primarily routine or non-routine, and cognitive or manual, using the mapping provided in Table A .1 of Cortes et al. (2014 ) . 3 Table A .1 in the Appendix shows the categorization I develop for Canadian NOC codes. For the purposes of the analysis, I differentiate only between routine and non-routine occupations (and not whether they are cognitive or manual), since that is the relevant dimension for determining whether the occupation can be automated readily. Canadian recessionary periods are defi ned using data from the Federal Reserve Bank of St. Louis that are based on OECD Composite Leading Indicators of GDP turning points in the growth cycle. 4 A recession is defi ned as starting in the month following the month containing the peak and ending in the month containing the trough (i.e., the trough is included but the peak is not). Throughout, I refer to recessions by the year of the trough. The data show three signifi cant recessions since 1987: the 1992 recession (from June 1989 to May 1992), the 2009 recession (from September 2007 to July 2009), and the 2016 recession (from October 2014 to June 2016). The last was a mild downturn not always considered an offi cial recession. Figure 1 presents monthly employment per capita for routine and non-routine jobs for the period 1987-2020, where recessionary periods are shaded in grey. The overall decline in routine jobs is apparent. Between January 1987 and December 2019, routine employment per capita fell by innovate and innovation incentives that increase during economic downturns. If productivity-enhancing activities (technological upgrading, reorganizations, and training) are disruptive to current production, we should observe more such activities during recessions when the opportunity costs are lower ( Hall 1991 ; Saint-Paul 1993 ; Aghion and Saint-Paul 1998 ) . This is sometimes referred to as the Schumpeterian opportunity cost hypothesis ( Barlevy 2004 ; Fabrizio and Tsolmon 2014 ) . Two alternative explanations for why automation increases during economic downturns are that managers shift attention from growth towards effi ciency due to an increased risk of fi rm failure ( Gibbons and Roberts 2012 ) and that downturns change the relative costs and benefi ts of layoffs ( Berger 2012 ) . The 1990s re-examination of the Schumpeterian view that recessions enable technological upgrading is not surprising, given that the economy had just experienced a recession and that it was undergoing the ICT revolution. The impact of this revolution on the labour force became the subject of enquiry, led by the seminal work of Autor, Levy, and Murnane (2003 ) . They and others (for example, Autor and Dorn 2013 ; Autor, Katz, and Kearney 2008 ; Goos and Manning 2007 ) document the phenomenon of job polarization, in which technology is displacing routine employment (be it cognitive or manual) in the middle of the skill and wage distribution. This greater understanding of the ICT revolution's impact on the labour market later allowed a deeper analysis of the Great Recession's impact on automation. The Great Recession accelerated not just the ICT revolution, but also the adoption of some emerging robotics and artifi cial intelligence technology. Kopytov, Roussanov, and Taschereau-Dumouchel (2018 ) develop a model to explain the greater share of investments going towards automation (software, information processing equipment, and robots) during the Great Recession and the associated job polarization. A number of empirical articles have also documented these trends. Jaimovich and Siu (2020 ; Henceforth JS) show that the process of job polarization accelerates during recessions. In particular, routine jobs experienced the biggest declines during each of the last three U.S. recessions and never recovered after the downturns, resulting in the much-publicized jobless recoveries. In fact, 88% of job losses in routine occupations between 1990 and 2017 occurred during the three recessions. Non-routine jobs, on the other hand, experienced only small declines and recovered quickly. Crucially, this dynamics was only present in the three recessions since the mid-1980s (which coincide with the beginning of the ICT revolution) and not the three previous ones. Hershbein and Kahn (2018 ) further show that in U.S. regions that were harder hit by the Great Recession, there was more investment in automation technology and a permanent increase in job skill requirements. Anghel, infectious new virus that we knew little about, and for which we were ill prepared, we took the unprecedented step of forcibly shutting down the economy in an effort to reduce its spread. However, even though the trigger for this recession was different, the shock of a generalized shutdown has produced a recession that cannot simply be reversed with a reopening. Hopes for a V-shaped recovery are misplaced. Demand is unlikely to rebound quickly. Depressed wages and profi ts, and in many cases increased debt loads, will be a drag on consumer and business spending. The May LFS reported that 22.5% of Canadian households expressed diffi culty meeting basic fi nancial obligations. While this is only slightly up from April (21.1%), it is likely to increase markedly when government supports are lifted. A further drag on spending is lingering uncertainty over the economy and potential new waves of COVID-19. On the supply side, output is also likely to experience a slow recovery. In the face of uncertainty, fi rms are unlikely to rush to rehire non-essential workers. In fact, if history is any guide, many fi rms will choose not to rehire them at all, taking this downturn as an opportunity to automate and reorganize their operations. Because reorganizations take time, they result in a temporary dip in fi rm output, which exacerbates the recession ( Kopytov et al. 2018 ). The COVID pandemic will also result in the downsizing or failure of many fi rms and the resources that they release will take time to be reallocated to other fi rms. For example, labour and capital are currently being reallocated from brick-and-mortar to online stores and services, but the reallocation cannot happen instantaneously, as online fi rms need time to expand and develop effi cient operations at a greater scale. Where the COVID recession may be significantly different from previous recessions is in the greater degree of economic transformation that it will induce. 5.4 percentage points, from 28.2% to 22.9%. Meanwhile, non-routine employment per capita increased by 5.6 percentage points over the same period. The fall in routine employment has not been a steady one. Rather, the fi gure suggests that routine employment drops precipitously during recessionary periods and otherwise is relatively stable (or even slightly increasing over the late 1990s and early 2000s). In fact, the recessionary periods, in spite of only representing 20% of the time, together account for a 6.5-percentage-point drop in routine per capita employment, representing 121% of the total 5.4-percentage-point drop between January 1987 and December 2019. If we compare this with the results of JS for the United States, they suggest that in Canada recessions play an even bigger role in accelerating automation and increasing productivity. Figure 2 examines each of the three most recent recessions more closely. Employment levels are expressed in terms of percentage deviations from employment at the trough of the recession and are shown for the three years prior to and following the trough. In all three recessions, non-routine jobs are unaffected, while routine jobs account for all of the job losses. More importantly, consistent with these jobs having been automated either directly or through reallocation, they never recover in the years that follow the recession. Overall, these trends are consistent with, and in fact stronger than, those found by JS for the United States in its three most recent recessions. The evidence strongly suggests that in Canada, too, recessions are crucial periods for automation and reallocation that increase long-run aggregate productivity. The COVID recession is different from earlier recessions in that it was largely self-imposed. In response to a highly decreasing worker-to-worker interactions. With the rise of interactive robots, it may be possible to intersperse these throughout operations in order to replace human-human interactions with human-robot ones (for example, handoffs of intermediate products between one station and the next). What is more, the way for fi rms to truly mitigate the risks of COVID to their operations, whether from workers getting sick or workers being ordered by government to stay home, is to fully automate their operations. All of this will amount to a signifi cant labour disruption. It is diffi cult to estimate exactly how many jobs will disappear permanently over the course of the COVID-19 crisis. Between February and May, seasonally adjusted employment losses totaled 1.27 million routine and 1.44 million non-routine jobs (17.3% and 12.1% of the precrisis levels, respectively). If this recession were like previous ones, we might expect these 1.27 million routine jobs to be lost permanently. However, these initial job losses are more related to the lockdown (which explains the also large reduction in non-routine employment) than to Superimposed onto the usual recessionary forces of automation are the COVID-specific health incentives to automate. We have already witnessed a tremendous willingness of organizations to respond to the challenges posed by COVID by adapting their operations. The April LFS showed that of the 12 million Canadians still working more than 50% of their usual hours, 5 million were mostly working from home (for two-thirds of them, this was a change from before the crisis). In spite of some reopening in May, the number held steady at 4.9 million, suggesting that some of these operational changes are likely to remain after the crisis. Moreover, this is likely to be but the beginning. The physical separation and digitization of tasks may mean that they are subsequently more easily offshored or automated with artifi cial intelligence. Yet automation is likely to be most prevalent, not among tasks and jobs that can be performed remotely, but rather for the majority that cannot. In these cases, organizations will have added incentives to automate parts of their processes in order to protect the health of their workers by Of course, just because there are incentives to automate a sector does not mean that it will be possible to do so. For instance, health care may experience high incentives to automate, but it is not clear that signifi cant automation is feasible, given current technologies. I construct a measure of feasibility of automation using the routine/non-routine categorization of occupations. To obtain an industry-level measure, I compute the fraction of workers in that industry who are in occupations that have been characterized as routine (once again using the 2016 Census Public Use Microdata File). Using the measures of automation incentives and feasibility, we can examine which industries are most likely to actually undergo signifi cant automation as a result of the COVID crisis. Figure 3 presents a scatterplot of the average physical proximity score ( y -axis) and the degree of routineness ( x -axis) for each of 19 NAICS industries. The average physical proximity scores of occupations within each industry range from 3 (slightly close, sharing an offi ce) to 4 (moderately close, at arm's length). Health care and social assistance is by far the industry requiring closest physical proximity, though education, accommodation and food services, and retail trade also exhibit high levels of average physical proximity. But on the dimension of automatability, health care and retail trade differ markedly. Industries towards the upper right of the fi gure are most likely to experience automation as a result of the COVID crisis. The most likely candidates are retail trade, but also construction, manufacturing, wholesale trade, and transportation. Not only do these industries have many routine tasks that are automatable, but also workers in those industries face signifi cant exposure to COVID, giving organizations added incentives to automate. Figure 4 presents a similar analysis, but with the extent of face-to-face discussions as a proxy for COVIDspecifi c incentives to automation. The results are similar. Retail trade continues to appear as the most poised for automation, but other good candidates are real estate, wholesale trade, construction, manufacturing, and mining and oil. Overall, the analysis suggests that while the COVID crisis is likely to result in signifi cant automation and reallocation that will increase aggregate productivity, the degree of transformation is likely to differ substantially across sectors. Government economic responses to the COVID crisis to date have focused on supporting businesses to ensure their survival, subsidizing wages so that fewer Canadians lose fi rm reorganizations as in previous recessions. It is thus likely that even routine jobs will rebound, yet just how far remains unclear. Adding to the diffi culty of estimating permanent job losses is the fact that fi rm-level automation and between-fi rm reallocation will continue shedding routine jobs over the course of the ongoing recession. History may be the best guide for estimating permanent job losses. The 1992 The , 2009 The , and 2016 recessions resulted in the loss of 484,000 (8.0% of prerecession levels), 330,000 (4.6%), and 359,000 (4.9%) routine jobs, respectively. Because of the relative severity of the current crisis, and due to the discussed COVID-specifi c incentives to automate, it is not unreasonable to expect routine job losses towards the upper range of 8.0%, as in the 1992 recession. This would imply the permanent loss of 586,000 routine jobs (or roughly half of the routine employment losses to date), but with it important gains in productivity. 5 In the following section, we consider which industries are likely to face the biggest health-related incentives to reorganize and which are most likely to be able to do so through automation. In addition to the usual recessionary incentives to automate, this COVID-19 crisis overlays health-related incentives that vary by industry. While health care workers face high levels of potential exposure, and hence the motivation for automation is high, agriculture and forestry workers face more limited exposure. To meaningfully discuss which industries face the strongest COVID-19 related incentives to automate, it is necessary to develop industry-level measures of potential worker exposure. To proxy for the degree of exposure, I use measures of occupational physical proximity and faceto-face discussions developed by the Vancouver School of Economics COVID-19 Research Program based on data from the Occupational Information Network. 6 Proximity refers to the extent to which the job requires close physical proximity to other people, and is ranked on a scale of 1 (do not work near other people) to 5 (work very close to other people). 7 Face-to-face refers to how often workers in the occupation have face-to-face discussions with individuals or teams, and is also ranked on a scale of 1 (never) to 5 (every day). 8 Of the two variables, proximity is likely the better measure of exposure, since some of what it captures is the necessity of performing physical tasks in person, whereas face-to-face discussions may more easily be moved online and not require automation to mitigate risks. Face-to-face is also less interesting in that it has relatively little useful variation; 97% of 4-digit occupations have scores between 4 (once a week or more) and 5 (every day). To construct the industry-level measure of exposure, for each industry I compute the weighted average of the level of exposure ( proximity or face-to-face ) faced by each The most signifi cant economic policies targeting fi rms are wage subsidies and increased access to credit. The Canada Emergency Wage Subsidy (CEWS) blunts the incentive of fi rms to automate by artifi cially reducing the cost of labour, and this at the exact time that opportunity costs for reorganizing operations are low. For example, while it might be profi table to automate the tasks being performed by a worker earning $20 per hour, it is unlikely to be profi table if the fi rm's cost is $5 per hour (after the 75% wage subsidy). In the best-case scenario, the productivityenhancing investment will merely be delayed; in the worst-case scenario, it will never occur, if by the time the subsidy is lifted, opportunity costs are again too high or the fi rm can no longer compete with fi rms in other jurisdictions that did engage in productivity upgrades. their jobs, and offering fi nancial support to Canadians who did lose their jobs. In short, the focus has been on maintaining the economy on life support, and rightly so. But in the coming months we must start shifting from a policy of survival and maintenance of the status quo toward one that embraces transformational change. The productivity upgrading of our economy will occur in two ways: fi rst, through the direct automation and reorganization by fi rms; second, through the reallocation of resources from less productive fi rms that are shrinking or outright failing to more productive growing ones. If our economy is to emerge from this crisis stronger, these mechanisms need to be encouraged, or at the very least not stymied. With this broader strategic objective in mind, policies that target businesses, industries, and individuals are considered in turn. The CEWS and the multitude of fi nancing and loans programs being offered to businesses on preferential terms all represent subsidies that have the potential to prop up ineffi cient fi rms, or ones that have been made obsolete by COVID-19. The crisis will cause large and persistent shifts in demand patterns (less demand for international travel, eat-in restaurants, in-person shopping) so that without subsidies many fi rms either would not survive the crisis or would shrink in size. These subsidies are costly, not just due to their burden on the taxpayer, but also because they may slow the reallocation of productive resources from less to more effi cient (or less to more adapted) fi rms, and thus lessen the productivityenhancing role of recessions. Such subsidies could also lengthen the time to recovery. To be sure, a delicate policy balance must be achieved between supporting good fi rms that might otherwise not survive the COVID shock and allowing the process of creative destruction to play its role. A 75% wage subsidy might have been warranted early in the crisis, but it almost surely will not be once the worst of the shock is behind us. As we move through this crisis, we must shift from policies that attempt to maintain the status quo to ones that foster innovation and reorganization, within both fi rms and industries. Such supports for innovation might be especially critical if fi rms had the will and know-how to reorganize, but not the fi nancial strength to do so, due to the crisis. In this regard, the recently announced $58 million Ontario Investment for Adoption of Digital Technologies is a good initiative that can hopefully pave the way for bigger ones like it. To go along with such investments, we also need to develop programs that help mobilize the tremendous expertise in robotics and artifi cial intelligence that resides in our universities and research centres for the benefi t of Canadian businesses. The federal government has announced numerous measures to support specifi c industries during the crisis. Many of these are aimed at the agriculture, agri-food, and fi sheries sector, which are strategically important for food security. In addition, special supports have also been put in place for the culture, heritage, and sport, air transportation, tourism, and higher education/research industries. However, the most signifi cant supports (CEWS and loans) apply equally across all industries. This is problematic because the impact of the COVID crisis, and in particular the relative costs and benefi ts of economic supports, differ widely by sector. Retail trade is undergoing a massive transformation. Individual organizations are embarking on major automation efforts. In physical stores this is spurred by tougher economic times, in online stores by a need to increase capacity, and in both cases by the necessity to limit the COVID exposure of workers. Yet the biggest component of the transformation is not within fi rms, but across them, in the form of a reallocation of market share and resources. Within this context, wage subsidies that attempt to maintain jobs are largely futile. In the case of expanding fi rms, they have little impact on employment and only serve as a transfer from taxpayers to (mostly large) fi rms. In the case of contracting fi rms, they only temporarily save jobs, slowing the process of reallocation and recovery. While retail is the clearest example of an industry where generous wage and loan supports seem counterproductive (it resides in the top right quadrant of Figure 3 ) , other industries such as construction, manufacturing, and transportation present similarly poor cases for government wage or loan subsidies. Instead, these industries could be offered supports that targeted R&D, innovation, and technological transformation. At the other extreme, industries in the top left quadrant (health care, education, accommodation and food services) present a good case for maintenance policies. Not only are they disproportionately impacted by COVID due to the high level of exposure of their employees, but also government support in these sectors is less likely to stifl e reorganizations. As we move forward and leave behind the worst of this crisis, which in economic terms was probably in mid-April, it is important that the current broad business supports are not renewed as they are set to expire. Instead, smaller, more targeted interventions should be deployed on an industry-by-industry basis. Often lost in the shuffl e of economic transformations are the individuals whose jobs have been affected. While investing in automation and facilitating reallocation are crucial to ensure continued aggregate productivity growth and an increasing standard of living for Canadians, they can carry a real human cost. Policies need to be in place to help individuals navigate the health crisis and ensuing economic transformation. The most signifi cant policy introduced to help individuals through the crisis is the Canada Emergency Response Benefi t (CERB). It has been crucial in ensuring that people can stay home and still fi nancially navigate the health crisis. It is not wholly without reproach, as it is well established that jobless benefi ts slow the reallocation of workers, because they have less incentive to fi nd new jobs ( Katz and Meyer 1990 ; Krueger and Mueller 2010 ) . But the current situation is one where working generates negative externalities in the form of increased infection rates. Regardless, the CERB is in place because individuals need it. Even as we move beyond the health crisis, some form of support will be required to navigate the upheaval of economic transformation. This should take the form of some continued fi nancial support (an extended CERB, employment insurance, or a different program altogether) but also of retraining programs for workers whose jobs have been automated or reallocated. This crisis may in fact be an opportunity to reimagine Canada's social safety net. COVID has generated a sense of solidarity and a newfound conviction of the important role of government. It has therefore created an opportunity to build institutions that will help Canadians not just through this current crisis, but also through future ones. We could consider, for example, a guaranteed basic income that would replace not just the CERB, but other current benefi ts, simplifying programs and ensuring that all Canadians enjoy a minimum standard of living regardless of circumstances. In drafting the appropriate response to the COVID pandemic and its aftermath, it is important to consider its many facets. This article makes the case that recessions, and this COVID crisis in particular, are periods of rapid automation and reorganization that are important to longrun aggregate productivity growth. Any policy should thus be seen within this context and evaluated for whether it is likely to facilitate or hinder technological change and reallocation. Put differently, it is time to start shifting from reactive policies designed to maintain the status quo to strategic policies designed to ensure that we emerge from the crisis stronger. For years, the AI and robotics revolution has appeared on the horizon as a bogeyman that will steal countless jobs. But to see it only as a challenge to be mitigated is to miss the point that automation will make Canadians richer and jobs more meaningful by removing menial tasks. It is a change that will come, and our choices are twofold: how quickly to adopt it, and what governance to put in place so that all Canadians benefi t. With respect to the fi rst choice, faster is better. We are at an infl ection point where the right policies can usher in rapid technological change and productivity increases. True, these will generate a period of signifi cant labour market upheaval as jobs are transformed, destroyed, and created. But now is the time to undergo such a transformation. The opportunity cost is lower because labour markets are already disrupted by the pandemic. Our choice is whether to pull off the Band-Aid slowly (and perhaps multiple times) or quickly, and emerge out the other end stronger. There is a further reason that now is the time to embrace change, and it relates to our second choice. This crisis has created a sense of solidarity that we are all in this together. It has also shaped a strong belief in the important role of government. This is an historic opportunity to rethink our social safety net, not just to help individuals weather the transition, but also to ensure that all Canadians share in the benefi ts of automation now, and for generations to come. It is hard to imagine a silver lining from this COVID pandemic, but with the right vision and leadership, a richer and fairer Canada may emerge. Virtues of Bad Times: Interaction between Productivity Growth and Economic Fluctuations Growth through Heterogeneous Innovations The Impact of the Great Recession on Employment Polarization in Spain The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market Trends in US Wage Inequality: Revising the Revisionists The Skill Content of Recent Technological Change: An Empirical Exploration Firm Dynamics and Productivity Growth: A Comparison of the Retail Trade and Manufacturing Sectors On the Timing of Innovation in Stochastic Schumpeterian Growth Models Covid-19 Is Also a Reallocation Shock Countercyclical Restructuring and Jobless Recoveries The Band Pass Filter The Micro and Macro of Disappearing Routine Jobs: A Flows Approach Gross Job Creation, Gross Job Destruction, and Employment Reallocation An Empirical Examination of the Procyclicality of R&D Investment and Innovation Aggregate Productivity Growth: Lessons from Microeconomic Evidence Market Selection, Reallocation, and Restructuring in the US Retail Trade Sector in the 1990s How Destructive Is Innovation ? Economic Theories of Incentives in Organizations Lousy and Lovely Jobs: The Rising Polarization of Work in Britain Recessions as Reorganizations Do Recessions Accelerate Routine-Biased Technological Change? Evidence from Vacancy Postings Job Polarization and Jobless Recoveries The Impact of the Potential Duration of Unemployment Benefi ts on the Duration of Unemployment Short-Run Pain, Long-Run Gain? Recessions and Technological Transformation Job Search and Unemployment Insurance: New Evidence from Time Use Data OECD Composite Leading Indicators: Reference Turning Points and Component Series Productivity Growth and the Structure of the Business Cycle Depressions: Can We Learn from Past Experience? Capitalism, Socialism and Democracy Vancouver School of Economics COVID-19 Nature of Work Risk Team Labor -Technology Substitution: Implications for Asset Pricing I would like to thank Robert Fay, Mikal Skuterud, and an anonymous guest editor for their comments. Kexuan Qu provided invaluable assistance in preparing the data. I acknowledge fi nancial support from the Region of Waterloo. Any errors and omissions are my own. 1 This includes employment of all non-institutionalized civilians aged 15 and over, not living on reserves or in extremely remote areas. Excluded groups represent less than 2% of the Canadian population aged 15 and over. The data are for 40 different occupational groups; all but two coincide with a single 2-digit NOC code (the other two are aggregates of several 2-digit codes). 2 I use the Stata implementation of the bandpass fi lter proposed by Christiano and Fitzgerald (2003 (2019). Information on the OECD data is available at OECD (n.d.). 5 Barrero, Bloom, and Davis (2020 ) estimate that in the United States 42% of COVID job losses will be permanent. 6 A user guide for their COVID risk/reward assessment tool, including a description of the variables used, is available at Vancouver School of Economics COVID-19 Nature of Work Risk Team (2020) . 7 The full scale for physical proximity is 1: do not work near other people, 2: work with others but not closely (e.g., private offi ce), 3: slightly close (e g., shared offi ce), 4: moderately close (at arm's length), 5: very close (nearly touching). 8 The full scale for face-to-face discussions is 1: never, 2: once a year or more, but not every month, 3: once a month or more, but not every week, 4: once a week or more, but not every day, and 5: every day.