Boza et al. (2023) compared the competitiveness of Spain’s and Chile’s olive oil industries using quantitative metrics and Porter’s diamond framework. Spanish producers were found to be generally more competitive due to factors such as stronger home demand and a better-established supply chain. Picture by John Cameron, Unsplash. Where is olive oil produced? Due to climate reasons, most of the world’s olive oil trees grow in the Mediterranean region. As a result, olive oil  production is also concentrated there – the olives must be processed soon after harvest (sometimes within 12 hours), which gives very little time for them to be exported and processed elsewhere. Among the producing countries, Spain is by far the largest: in 2023/24, it produced around a third of the world’s olive oil. Chile, in contrast, made up just 1% of the global output. (See Figure 1.) Figure 1:  Spain is by far the world’s largest olive oil producer Source: World population review. Data manipulated by Bonsai Economics. A quantitative assessment of international competitiveness There is no single way to measure a country’s advantage or disadvantage in selling its products in the international market. Hence, Boza et al (2023) used three different metrics: international market shares (IMS), and the revealed comparative advantage (RCA) and trade competitiveness (TC) indices. We will just focus on the IMS metric as it is the easiest one to interpret.   The IMS metric hinges on the notion that a country that accounts for a larger share of the world’s exports of a product must be generally more competitive than one with a smaller share. In other words, if country A accounts for 40% of all olive oil exports, and country B for 5%, then country A must have some overall relative advantage in selling its product internationally. Unsurprisingly given the relative size of their overall production, the IMS metric suggests that Spain’s olive oil industry is much more competitive internationally than Chile’s. (See Figure 2.) However, it does not help explain why . Figure 2:  Spain has often accounted for over 40% of all olive oil exports Source: Boza et al. (2023), Bonsai Economics. Based on data from WITS-World Bank (2023). Note: IMS for Spanish olive oil = Spanish olive oil exports / global olive oil exports. Porter’s Diamond Framework To understand why countries’ relative competitiveness might differ, Boza et al. (2023) relied on Porter’s diamond model. This framework was first developed by Michael Porter in 1990, and it distinguishes between four interrelated drivers of competitiveness: Factor conditions. These include physical resources, labour, and general know-how. Demand conditions. They refer to the domestic  need for the product (e.g. demand for olive oil within  Chile), as the home market is seen as a stepping stone to the international one. Related and supporting industries. Firm strategy, structure and rivalry. It refers to the ways companies are organised, managed, and compete with each other. Government policy was also included as a factor that could influence any of the other conditions. (See Figure 3.) Figure 3:  Porter’s Diamond Model focuses on four key drivers of competitiveness Source: Porter (1990), Bonsai Economics. How does Porter’s framework apply to the olive oil industry? Boza et al. (2023) gathered information on Porter’s factors through a series of semi-structured interviews with market participants from Spain and Chile. These allowed for various cross-country comparisons, out of which we will highlight four. First, the structure of the two industries was found to be starkly different. In Spain, most production happened through cooperatives of small and medium-sized farmers. But in Chile, it was larger companies that dominated. It is unclear whether one of the two structures is generally more advantageous. Second, home demand conditions were reported to be much stronger in Spain than in Chile. This result is unsurprising given the more central role of olive oil in the Spanish diet. In fact, an interviewee stated that annual per capita olive oil consumption in Spain was around 15 litres, whereas in Chile just 0.5. According to Porter’s framework, this may translate into a competitive advantage for Spanish producers. Third, Spanish producers reported that they generally had better access to production inputs such as machinery and spare parts. While many supplying companies have branches in Spain, in Chile there are just a few local suppliers, and most inputs must be imported from abroad. For instance, almost all Chilean interviewees highlighted the challenge of having only two suppliers of glass containers, which prioritise the wine industry over the olive oil sector. This was reported to have often limited Chile’s olive oil production capacity. Fourth, olive oil production is highly subsidised in Spain, but not in Chile. The EU’s Common Agricultual Policy (CAP), for example, provides subsidies for Spanish farmers that can represent 20-25% of the overall price. In Chile, most producers are not eligible for support, as it only covers small-scale businesses. Conclusion Overall, Boza et al. (2023) highlighted various reasons for the relative success of the Spanish olive oil sector, such as stronger home demand and government subsidies. However, the relative importance of each factor remains unclear.

Jovanovic et al (2005) analyzed the impact of two general-purpose technologies (GPTs) on the US economy: electricity and information technology. They found that there were considerable time lags between each technology’s invention and widespread adoption. As a result, it took over 20 years until each GPT had a material impact on economy-wide productivity. Picture by Alexandre Debieve, Unsplash What are general-purpose technologies? General-purpose technologies (GPTs) are those technologies that transform both household life and business conduct. More specifically, according to Bresnahan and Trajtenberg (1996) , a GPT should have the following three characteristics: Pervasiveness:  the technology should spread to most sectors. Improvement:  it should get better over time, lowering the costs of its users. Innovation spawning:  it should make it easier to invent and produce new products or processes. Some examples that meet these criteria are steam, internal combustion, electricity, and information technology. In this short article, we will be focusing on the latter two. When did each GPT spread? Jovanovic et al (2005) allocated each GPT to its own respective era, which spanned from its initial adoption to its establishment as a widespread technology. Each period’s start and endpoints were determined by each GPT’s adoption rate: an era began when the GPT achieved a 1% diffusion in the median sector, and it ended once the diffusion curve flattened. As a result, they argued the Electrification era started in 1894, with the completion of the first hydroelectric facility at Niagara Falls, and that it ended in 1930. For the Information Technology (IT) era, they claimed it began in 1971 with the invention of Intel’s ’4004’ microprocessor, the key component of the personal computer. The IT diffusion curve hadn’t flattened at the time of their paper, so they viewed the era as ongoing. There were lags in each technology’s adoption… A key message from the Jovanovic et al (2005) paper was that there can be considerable time lags between the invention of a GPT and its widespread usage. In fact, both for electric appliances and personal computers, it took almost 30 years from their invention to reach 50% of all households (see Figure 1, below).   Figure 1: It took almost 30 years for appliances and PCs to reach 50% of households Sources: Jovanovic and Rousseau (2005). What’s more, in manufacturing, it took many businesses over 20 years they complete the electrification of their processes (see Figure 2, below). Figure 2: The electrification of manufacturing was a lengthy process Sources: DuBoff (1964), Jovanovic and Rousseau (2005). Jovanovic et al (2005) , offered various explanations for these lags. For one, it took time until the technologies were cheap enough to be accessed by the wider public. Affordable PCs, for example, only came out in the 1980s, when the technology was some 15 years old. And for electrification, the durability of old manufacturing plants which relied on water and steam, often didn’t make it economical to replace them. For this reason, as noted by David (1989) , it was expanding  industries that tended to electrify first. In the early twentieth century, these were tobacco, fabricated metals, transportation equipment, and electrical machinery. …productivity gains were also late to materialize These lags form a part of the reason why economy-wide productivity only accelerated 20 years after the invention of each GPT (see Figure 3, below). But they were not the only reasons. As highlighted by David (1989), GPTs may lead to qualitative changes not captured by productivity statistics. Shorter traveling times for workers via electric trams are such an example. What’s more, due to the revolutionary nature of such technologies, their initial adoption may have an exploratory character that doesn’t take advantage of all the available efficiencies.   Figure 3: After the invention of each GPT, it took ~20 years for productivity growth to accelerate Sources: Federal Reserve, Bonsai Economics. Conclusion Looking ahead, in the case of artificial intelligence, it would be unwise to assume that its adoption must follow a similar pattern to previous GPTs. However, these past experiences may help us be open to the possibility that new technologies may take some time until their revolutionary impact is fully felt. And once this occurs, productivity growth may accelerate considerably.

Bardazzi et al (2023) argue that the complex nature of energy poverty prevents it from being captured by a single variable. For this reason, they recommend to policymakers that they monitor energy poverty by using a dashboard of metrics. Picture by William Warby, Unsplash. What is energy poverty? A person is said to experience energy poverty (EP) if they are unable to access the energy services that are considered essential for a satisfactory standard of living and health. Although this qualitative definition is widely accepted, its quantitative equivalent is not. In the UK alone, for example, England, Scotland, Wales and Northern Ireland all have their own ways of measuring energy poverty. Furthermore, confusingly, the conclusions reached by using each metric don’t necessarily align with each other. How can we measure it? According to Bardazzi et al (2023), EP metrics can be categorized as objective, consensual, or multi-dimensional. Objective metrics identify EP households by using data on households’ energy and total spending. The ones emphasized (and criticized) in the paper are mostly the following: Ten Per Cent Rule. This was the first EP metric to emerge, and it was developed in the UK after the 1970s oil shock, when rocketing oil prices forced vulnerable families to spend an important part of their budget on energy bills. According to the indicator, a household experiences EP if it spends over 10% of its disposable income on energy. Although the metric has the advantage of being easy to communicate, Bardazzi et al (2023) highlight a couple of shortcomings. For one, the 10% threshold was derived using empirical data for the UK back in the 1980s, so it may no longer be a good benchmark. What’s more, it doesn’t capture those EP households which radically lower energy usage to afford more important necessities such as food. M/2 indicator. According to this, households are likely to experience EP if their energy expenditure is below half the median of the reference population. This metric isn’t perfect either; low energy expenditure could be driven by factors such as high energy efficient dwellings, or specific rent contracts which include utility bills. And by capturing ‘relative’ expenditure, M/2 is arguably more of a measure of energy spending inequality rather than absolute poverty. Low Income High Cost (LIHC) metric. This identifies EP when a household faces both high energy costs compared to the target population, and its residual disposable income (net of energy expenditure) is below the poverty line. Similarly to the ‘Ten Per Cent Rule’, the LIHC fails to capture those EP households which choose to lower their energy usage drastically. Modified LIHC (M-LIHC) metric. This variation of the LIHC was proposed by Faiella and Lavecchia for Italy in 2014, and it expanded the LIHC EP identification criteria to include those households that have zero energy expenditure and are below the poverty line. Although an improvement from the LIHC, Bardazzi et al (2023) argue that the zero-expenditure threshold could be overly strict. Consensual indicators, in contrast with objective metrics, do not use data on households’ expenditures, but instead they rely on relevant self-reported survey information. Two common examples are presented below: Survey results on perceived ability to adequately heat the house in winter. An issue with this indicator is that people’s answers may be affected by cultural and demographic characteristics over what is considered an ‘adequate’ temperature. Deller et al (2021), for example, found that older households are less likely than younger ones to self-identify as unable to afford adequate warmth. Self-reported information on whether a household was late to pay their energy bill. Similar to other metrics, this is unlikely to capture those EP households that avoid using energy services. Lastly, multi-dimensional EP metrics use weighted averages to merge various objective and consensual metrics into a single indicator, to combine their insights. However, such indices can be hard to interpret, making them difficult to use. Are energy poverty metrics interchangeable? Bardazzi et al (2023) find that EP metrics don’t necessarily align when it comes to comparing the prevalence of energy poverty across countries. According to surveys on perceived ability to heat the home, for example, Italy had the 6th highest EP rate in the EU in 2018. In contrast, self-reported information on late payments of energy bills ranked Italy as low as 21st place (see Figure 1). Similarly, the M/2 metric indicates that Sweden had the highest EP rate in the EU in 2010, while the M-LIHC indicator positioned it at 20th place (see Figure 2). Figure 1: Consensual metrics ranked Italy at differing positions for energy poverty. Data: EU Household Budget Survey, Bardazzi et al (2023). Figure 2: Objective metrics positioned Sweden at 1st and 20th place for EP. Data: Statistics on Income and Living Conditions (EU-SILC data), Bardazzi et al (2023). Conclusion There is evidence of inconsistencies across indicators, at least when it comes to the level of EP across countries. At the same time, according to Bardazzi et al (2023), no metric is clearly superior to the others. As a result, they suggest that a dashboard approach of indicators is the best way to track trends in energy poverty.

Maniatis et al (2023) discuss the rise of energy communities in Greece since they were first introduced by legislation in 2018. The Agrinio and the Minoan energy communities are presented to be among the most successful ones. Picture by Shane Rounce, Unsplash Communities and the Green Economy The paper argues that the more decentralized nature of green energy, as opposed to fossil fuels, can empower the role of communities in energy markets. With renewable energy sources such as solar and wind, it is feasible for many households and small businesses to develop their own power plants. As a result, communities can be important actors in helping decarbonize energy markets and tackle energy poverty. Such communities exist across Europe. In France, for example, Enercoop is an energy cooperative that has over 400 production sites. And in the UK, Energy4All has been offering financial and management services to such groups since 2002. The growing role of energy communities is also evident from some recent legislation by the European Commission: in April 2022, they launched the Energy Communities Repository, which aims to support the development of such collectives. The case of Greece In Greece, energy communities have a short history, as they were only introduced in legislation in early 2018 (Law 4513/2018). According to this, members of energy communities can be citizens, legal entities, and regional and local authorities. Such collectives are free to consume or trade the energy they produce, and they may choose to support vulnerable consumers within their area of responsibility. As they are meant to benefit their own locality, 50% + 1 of community members need to be related to the location of each of their projects. According to some surveys referenced by Maniatis et al, Greece has over 400 energy communities, mostly located in Central Macedonia. The majority of these (~75%) have 10-20 members, and they have implemented a total of over 430 projects (mostly photovoltaic). The paper chooses to highlight two for their successes: Agrinio Energy Community (AEC). It was formed by the Agricultural Cooperative of Agrinio (ACA), which is a collective dating back to the 1930s, and has traditionally been active in the agricultural sector. The cooperative has ~300 members, who are mostly tobacco and oil producers from the provinces of Trichonis, Xiromero, and Valtou. The cooperative manages ~15 energy communities, including 10 wind farms with 168 MW capacity and seven solar farms with 126 MW capacity. Minoan Energy Community (MEC). Established in 2019, this is the first energy community to be established on the island of Crete. The collective has over 200 members, including municipalities, small businesses, and individual citizens, and it has commissioned a photovoltaic plant with a capacity of 405 kilowatt peak (kWp). Furthermore, it plans to cover the energy needs of low-income households in the area recently suffering from earthquakes (Arkalochori earthquake of 2021) and other natural disasters. In the words of one member, as cited by Maniatis et al, “the community is all about solidarity and goodwill because we cannot always rely on outside support”. Conclusion Admittedly, with just six years of history, energy communities in Greece are still in an early stage, with many citizens and entities probably still unaware of the new opportunities they have. What’s more, Maniatis et al argue that there is still significant scope for the relevant legislation and incentive schemes to improve and help the formation and success of such initiatives. But given positive examples such as the AEC and MEC, there seems to be space for optimism.

In his 1950 paper ‘Vertical Integration and Antitrust Policy’, Spengler posited that vertical mergers could lead to lower consumer prices, by addressing the problem of ‘double marginalisation’. In essence, he argued that vertical integration could rid industrial value chains of ‘costly middlemen’, lowering production costs and reducing final prices. Picture by Samson, Unsplash What are the different types of value chain integration? Industrial economics broadly distinguishes between two types of value chain integration. First, horizontal integration, which involves the merging of companies operating in the same part of a value chain (e.g. merger of two copper rod producers). And second, vertical integration, which refers to the combination of firms operating in different stages of production (e.g. merger of a rod producer and a cable manufacturer). A key concern of antitrust policy is that mergers that result in either type of integration, have the potential to undermine market competition, resulting in higher prices and other socially adverse outcomes. Spengler’s contribution was to highlight that, in the aftermath of extensive horizontal integration, vertical integration could actually lower prices. The problem of double marginalisation Consider a scenario in which horizontal integration within the copper rod market has weakened competition between rod producers. As a result, rod manufacturers demand an above-competitive price for their rods, pushing up the costs of cable producers. A portion of these heightened costs is then passed on to cable consumers, resulting in higher final prices. Under these market conditions, imagine the impact of a merger between a rod producer and a cable manufacturer. The cable manufacturer no longer needs to pay the ‘excessive’ surcharge to the rod producer. Consequently, the cable manufacturer’s production costs drop and a portion of these cost savings are transferred to cable consumers via lower prices. In other words, optimizing profit across both stages of production simultaneously, as opposed to independently (‘double marginalisation’), could lower prices for consumers. Conclusion Spengler’s paper marked a significant milestone in the evolution of antitrust practice. But it shouldn’t be interpreted to imply that vertical integration never harms consumers. Accordingly, authorities should evaluate the market impact of vertical mergers on a case-by-case basis.

Schmitt (2017) found evidence that multi-market contact in the US hospital industry has led to higher hospital prices. This suggests that even if future hospital consolidation only has a marginal impact on within-market concentration, it may still produce higher prices. Picture by National Cancer Institute, Unsplash How consolidation reshaped the US hospital industry The US hospital industry has witnessed significant consolidation since the start of the 21st century. According to the American Hospital Association, from 2000 to 2010, for example, approximately 60 hospital mergers occurred annually; this figure rose to nearly 100 mergers per year between 2011 and 2014. The rapid consolidation of the US hospital industry has affected hospital competition in two major ways: Increased within-market concentration: Within each hospital market — typically defined as a geographic region — there has been a drop in the number of hospitals competing against each other. The proportion of hospital referral regions (HRRs) exhibiting high concentration, for example, as defined by the US Horizontal Merger Guidelines (2010), has risen from 43% in 2000 to 58% in 2012. Rise in multi-market contact: As multihospital organisations became more prominent, hospital companies increasingly competed in multiple geographic markets. In 2012, for example, 29% of HRRs had no multi-market contact at all, compared to over 40% in 2000. Schmitt (2017) specifically focused on the impact of multi-market contact on hospital prices, and prices were estimated using data on average hospital revenues. Multi-market contact as a hindrance to competition Multi-market contact can hinder competition because it can foster collusion between firms. After all, when firms compete in multiple markets, they may hesitate to engage in vigorous competition in a specific market, out of fear that their rival will retaliate by intensifying competition in all the other markets they compete in. Consequently, even in the absence of an explicit (and illegal) agreement between competitors, firms may choose to keep prices higher than they would have under vigorous competition. There already exists empirical evidence that multi-market contact can lead to higher prices. Evans and Kessides (1994), for example, found that airline fares are significantly higher on routes where there exist carriers that have contracts on several routes. But not all arguments agree that multi-market contact hinders competition; Berheim and Whinston (1990), for instance, showed that when firms and markets are perfectly symmetric, multi-market contact doesn’t change the incentives to collude. As a result, whether a higher degree of multi-market contact in the US hospital industry has produced less competitive prices, is ultimately an empirical question. A simplified overview of the paper’s methodology A difficulty with estimating the price impact of multi-market contact is that companies may choose to acquire the hospitals which they expect to have the greatest price rises in the future. Consequently, elevated post-merger prices in these hospitals could be driven by hidden market-specific drivers different from multi-market contact. Unless addressed, such an endogeneity problem can prevent the estimation of the causal effect of multi-market contact on hospital prices. Schmitt (2017) dealt with the endogeneity issue by only estimating the price effect of multi-market contact on hospitals that were not themselves part of the merger, or within the market where the merger happened, but still belonged to the merged multihospital system. This way, the paper excluded from its analysis any a priori company expectations about the merged hospitals’ future prices. And in the simplest version of their model, they used the fixed effects regression below, where λ attempted to capture the causal impact of multi-market contact on hospital prices. Some evidence of anti-competitive pricing In the simplified version of their model, Schmitt (2017) estimated that multi-market contact over the 2000-2010 period raised prices by almost 7%, and the results were supported by various robustness checks. This suggests that regulators should not only focus on the impact of hospital mergers on within-market concentration but also on its impact on cross-market hospital interactions.

In a 2001 paper, Robert Pitofsky, former Commissioner of the US Federal Trade Commission, examined the challenges faced by antitrust authorities when dealing with the “New Economy”. Pitofsky identified three common characteristics of the New Economy which should be considered by competition policy: a high reliance on intellectual property, a strong focus on product innovation, and market share volatility. Picture by Joshua Sortino, Unsplash What is the “New Economy”? As discussed by Pohjola (2002), the term “New Economy” first emerged during the 1990s, as many became convinced that the world economy was undergoing two fundamental structural transformations: the growing globalisation of business, and a revolution in information and communication technologies (ICT). These forces were expected to lead to a superior economic structure, which the business press described as the “New Economy”. Pitofsky (2001) focused on the ICT dimension of the “New Economy”, to which he also added the biotechnology sector. Some key characteristics of the New Economy Pitofsky (2001) contends that the New Economy, though not fundamentally distinct from the “old” one, often exhibits three key characteristics that competition authorities should consider: Intellectual property reliance: In the New Economy, products are frequently built on innovative ideas that require significant initial investments but have low reproduction costs for individual items. To incentivise firms to make these investments, patents and copyrights are often used to prevent competitors from immediately copying the innovations. But this protection comes at the expense of weakened market competition, as innovation is safeguarded by limiting access to others. Emphasis on innovation: According to Pitofsky (2001), consumers in the New Economy often benefit more when firms focus on creating the best and newest product, rather than when they solely focus on cost minimisation and quantity maximisation. In the words of the paper, “the key is not so much who can produce the most widgets at the lowest cost, but rather who can be the first to design, protect with intellectual property rights, and bring to market a new and improved widget”. As a result, economic models that solely emphasise price and quantity competition have limited applicability in the New Economy. Uncertain durability of market power: the paper argues that there was a consensus at the time that market power in the New Economy would frequently be short-lived due to rapid innovation and relatively easy market entry. However, the author mentioned some concerns with this view, highlighting that network effects and high sunk costs (e.g. advertising) can often increase the barriers to entry significantly. Conclusion Over two decades after Pitofsky wrote his paper, it’s hard to deny the New Economy’s continued reliance on innovation and intellectual property rights. But the extent to which market power durability is more uncertain in the New Economy than elsewhere remains unclear. This is especially true considering the growing concerns among several competition authorities in recent years regarding the power of Big Tech.

In their seminal 2010 paper, Farrell and Shapiro introduced a novel metric for the evaluation of unilateral price effects in horizontal mergers, which they called Upward Pricing Pressure (UPP). Since then, the UPP has emerged as a viable alternative to conventional concentration indices such as the Herfindahl-Hirschman Index (HHI). Picture by Unsplash Why are mergers screened? One of the primary concerns in company mergers revolves around the potential amplification of market power, wherein the merged entities, no longer competing, decide to raise prices above pre-merger levels. Meanwhile, antitrust authorities lack the resources to examine in detail all the proposed mergers for anticompetitive effects. As a result, in practice, simple preliminary tests are commonly used to identify mergers with a higher likelihood of harming consumers. These mergers are then typically subjected to more detailed examinations, in which authorities assess the validity of any competition concerns. The Herfindahl-Hirschman Index (HHI) A commonly used indicator for merger screening is the HHI, which is calculated by adding the squares of the market share of all companies in a particular market, scaled by 100. The precise formula is presented below: where si is the share of the ith firm. The HHI is a measurement of market concentration, and it falls with the number of effective firms in the market. In qualitative terms, according to the US Horizontal Merger Guidelines (2010), an HHI above 2,500 implies that a market is highly concentrated, between 1,500 and 2,500 that it’s moderately concentrated, and below 1,500 that it’s unconcentrated (see Figure 1, below). Figure 1: The HHI falls with the number of effective firms in a market. Data: Bonsai Economics, US Department of Justice Horizontal Merger Guidelines (2010) By assuming that market concentration indicates market power, competition authorities investigate mergers that lead to significantly more concentrated markets. In the US, for example, a rise of more than 100 points in the HHI in moderately and highly concentrated markets is flagged as a potential risk to competition, and it can trigger authorities to scrutinise further the proposed merger. The market definition issue In their paper, Farrell and Shapiro (2010) highlight a notable limitation of market concentration measures such as the HHI – their heavy dependence on the definition of the relevant market (you need a market definition to derive the company market shares). To substantiate their criticism, the authors delve into the case of the proposed merger between Whole Foods and Wild Oats in 2007, in which both companies were grocery chains specialising in natural and organic food. In this case, the Federal Trade Commission (FTC) and the grocery stores used different market definitions, which led to different views on the impact of the merger. The FTC insisted on defining the relevant market as “premium natural/ organic supermarkets”, which suggested the merger would substantially increase market concentration in some locales. In contrast, the grocery stores asserted that their customers also shopped at traditional supermarkets, which should be included in the relevant market. Under this latter definition, the anticipated increase in concentration would be minimal. According to the authors, the court was unable to find a robust way to choose between the two; this suggested to them that concentration measures weren’t always appropriate for the evaluation of mergers in markets with differentiated products. The Upward Pricing Pressure (UPP) alternative Farrell and Shapiro (2010) introduced an interesting alternative to traditional market concentration measures – Upward Pricing Pressure (UPP). Departing from reliance on specific market definitions, the UPP assesses the price implications of mergers by considering two opposing forces: The upward pricing pressure from the elimination of competition between merging firms. It is quantified by evaluating the degree to which a price increase in the merging firm’s product would simply divert consumers to the other one’s product. The downward pricing pressure from the efficiencies generated by the merger. It is typically estimated by reference to some other analogous merger. Assuming that two firms compete on price with differentiated products, the UPP for firm 1’s product can be calculated using the following formula: where D12 denotes the diversion ratio from firm 1’s product to firm 2’s, ‘P’ and ‘C’ denote the price and cost of each firm’s product, and ‘E1’ is the efficiency gain of firm 1 because of the merger. Using this formula, a positive UPP indicates that the merger’s adverse impact on competition may outweigh any efficiency gains, necessitating a more thorough examination of the merger. Conclusion Overall, the UPP presents a robust alternative to traditional market concentration measures, particularly in situations where determining the appropriate market boundaries is difficult. Nevertheless, as with the HHI, the UPP should not be regarded as conclusive evidence that a merger would harm competition. Instead, Farrell and Shapiro (2010) propose that the UPP should be primarily used as a screening tool, pinpointing those mergers which warrant closer examination by authorities.

There is a spectre haunting our lives, and that is chaos theory. It’s all around us, but we hardly see it. Understanding it gets to the heart of some of the most interesting debates and dichotomies in academia: humanities versus social science, inductive versus deductive reasoning, classical versus romantic, hippies versus yuppies. Picture by Saheb Zaidi, Unsplash In the 17th century, Francis Bacon published the New Organon, ushering in the modern age of scientific discovery. We still struggle to know the true causes of things, but we’re a far way off from thinking the universe was made of 4 elements or defining humans as ‘featherless bipeds’ (cue Diogenes plucking the feathers off a chicken and chucking it into the forum, taunting Plato, ‘Behold, a man!’). But epistemologically, we have to ask ourselves how advanced we really are: modern medicine was mostly useless until the 1930s; we live in an age where quackery and misuse of statistics and information is ripe. This is because, as is relatively common knowledge now, humans often act quite irrationally. This simple truism, I posit, is partly why we have so much trouble not just predicting future actions, but even explaining past ones as well. One way we can think about the limits of our knowledge is the application of ‘chaos theory’, which behind its cool name lies a deceptively powerful concept that most social scientists would do well to heed. At its heart, chaos theory argues ‘that some social systems are characterized by ‘chaotic’ behaviour which precludes the possibility of predicting the future of that system’. Most famously popularized by notions such as the ‘butterfly effect’, whereby even the smallest event (e.g. a butterfly flapping its wings over Cambridge) can have a massive impact (a hurricane over Kowloon). This idea has its roots in another crucial development of statistics (themselves a pillar of any social science and prediction), namely ‘Laplace’s Demon’, which pondered whether if one knew the exact location and direction of every atom in the universe, could one predict the future? Chaos theory says no, for both practical and epistemological reasons. Why can’t we know the future? Often it is a measurement problem. We can’t measure every atom, let alone to a degree that would allow us to predict their future. An example to highlight this is in weather forecasting, and it was when chaos theory was first ‘discovered’. Even in physical systems (with no sentient beings to model, who are fickle to say the least), minutia matters. In the computer-infancy 1980s, when Lorenz was modelling climate patterns and wanted to rerun an experiment through his computer, he manually reinputted his numbers into the machine. However, he only put in 8 decimal places (this is all the printer printed off for him). This change is infinitesimal – a millionth of a degree – but despite having changed no parameters, the lack of these miniscule decimals was enough to make the results completely different. How can a few decimal places affect the outcome so radically? In a nutshell, this is the effect of chaos on physical systems. But how does this work in social science? Chaos in social systems has two key definitions: It detectable by its ‘highly iterative, recursive, or dynamic structures that change over time’. It demonstrates ‘highly discontinuous behaviour in the system’ – this is sometimes also described as being ‘fractal’. One famous example of this are the so-called Mandelbrot Sets, in which for 'a substantial portion of the values of the [input], extremely close neighbours behave in opposite manners’. Or, in layman’s terms, Mandelbrot Sets are equations which demonstrate stability and instability in a pattern that is completely unpredictable: inputs so close and similar can generate wildly different outputs. This unpredictable relationship is the mathematical (and natural) manifestation of chaos. And this is why chaos theory is so important to social sciences: one can build complex models, with multivariate prediction algorithms – and end up completely wrong. What else can be tried to remedy this? What about increasing the number of actors (‘increasing our n-value’, in the parlance of statisticians) – will this help us forecast with more certainty? Sadly, also no: ‘even in a study with a large number of actors, there can be a very large number of cases where arbitrarily similar actors can display radically different behaviours.’ Even our fanciest mathematical tricks – statistics, linear and non-linear regressions, or structural equations – are woefully unhelpful when studying chaotic systems. So, do we just give up? Yes. And no. It is true that ‘many social systems reflect the central components of chaos theory.’ Understanding chaos theory helps direct us towards where we can predict – and helps us avoid where we can’t. When studying chaotic systems (as described above), traditional social science often relies on a few worn-out tricks: a) searching (in vain) for some omitted variable that will explain the chaos, b) thinking their measurement is the problem, or c) worrying that the stochastic part has overwhelmed the patterned part (and not seeing that randomness is the pattern, essentially). Like the Bed of Procrustes, we should be wary of fitting chaotic systems onto models which rely on tools for studying non-chaotic systems. This blog post does not claim to have the solution but knowing what we’re doing wrong – i.e. using cross-sectional research designs to fallaciously predict the chaotic world around us – is as good a start as any.

Knotek (2007) examined whether the statistical relationship of real output growth with the unemployment rate, as captured by Arthur Okun’s simple rules of thumb, has been stable over time in the US. The paper showed that the relationship has varied over the business cycle, and more generally across history. Picture by the New York Public Library, Unsplash What is Okun’s law? ‘Okun’s law’ refers to a set of statistical relationships that describe the inverse relationship between changes in the unemployment rate and real economic output. Formulated by Arthur Okun in the early 1960s, it originally came in the form of two main versions, which are presented in Figure 1 below. Figure 1: Okun’s law originally had two main general forms Sources: Knotek (2007), Bonsai Economics. Out of the two, Knotek (2007) argued that the ‘gap’ version is generally harder to use. After all, it relies on estimates of how much the economy would produce under conditions of maximum sustainable employment (‘potential output’), which can vary depending on the methodology used to derive it. The ‘difference’ version, in contrast, simply associates the unemployment rate with contemporaneous real economic output growth. How stable is Okun’s coefficient? Focusing on the ‘difference’ version, Knotek (2007) made two interesting observations about the historical stability of Okun’s coefficient (β from Figure 1 above). First, Okun’s coefficient has generally tended to rise in periods of economic expansion. This is illustrated in Figure 2 below, which uses a five-year rolling sample to estimate the coefficient over time. In all four occurrences when the sample didn’t include recession periods, Okun’s coefficient estimate rose significantly. Figure 2: Okun’s coefficient (β) has generally tended to rise in economic expansions Source: Knotek (2007) Second, in more recent periods, while the contemporaneous relationship between unemployment and real output has weakened, the relationship of unemployment with lagged real output has strengthened (see Figure 3, below). This finding favours forms of the ‘difference’ version which also include lags for economic output growth. Figure 3: Lag coefficients are more important for more recent data Source: Knotek (2007) Conclusion The upshot is that Okun’s law, contrary to the connotations of the word “law”, is only a rule of thumb, and the exact statistical relationship can vary over time and according to broader economic conditions. As a result, if one were to forecast the unemployment rate via Okun’s law, it would probably be more appropriate to use a more general form that captures its changing nature.

Alfred Marshall’s derivation of ‘consumer surplus’ in his book ‘Principles of Economics’, was one of the first attempts by economists to measure the effect of prices on people’s welfare. It also relied on a set of unrealistic assumptions, which raised much controversy among economists back in the time. Picture of Alfred Marshall Marshallian consumer surplus – what is it? When Marshall first published his book ‘Principles of Economics’ in 1890, one of his aims was to provide a general approach for measuring how much people ‘truly’ value the goods they purchase. After all, the price of a good itself often isn’t a great measure of value. I could buy a car for £15,000, for example, while I would still get it for £20,000. In this scenario, according to Marshall’s thinking, not only am I buying the car, but I also earn another £5,000 of “surplus satisfaction”, which is my ‘consumer surplus’. In other words, my consumer surplus is the extra money I saved by paying less than how much I ‘truly’ value the car. Marshall develops his argument further by incorporating in his analysis a consumer’s demand function. Let’s assume, for example, that one chooses the number of plain T-shirts they buy according to Figure 1 below (i.e. they buy one T-shirt if it costs £15, two if they cost £12 each, three if they cost £9 each, etc.). And let’s assume a T-shirt costs £12, which means they would buy two T-shirts. To estimate the consumer’s consumer surplus in this scenario, Marshall roughly uses the following method: We already know they would buy one T-shirt even if it cost £15, which implies they value it at £15. Given that they paid £12 for this T-shirt, this suggests that they have a ‘surplus satisfaction’ of £3 from the purchase of the first T-shirt (£15 - £12 = £3). We also know they wouldn’t buy a second T-shirt if the price was £15, which means they value it less than £15. After all, the more T-shirts they have already, the less they value any new ones (diminishing marginal utility of consumption). And since they would buy a second T-shirt if it cost £12, then they value it at £12, which suggests they get no consumer surplus from this purchase (£12 - £12 = £0). Consequently, their total consumer surplus from buying the two T-shirts is £3 (£3 + £0). Figure 1: A consumer’s demand function for T-shirts Illustration by Bonsai Economics. The constant marginal utility of money assumption While Marshall’s consumer surplus can initially impress with the simplicity of its logic, it relies on a variety of assumptions. One of the most contentious ones is that people’s wealth doesn’t affect the amount of goods they purchase (or, in formal terms, that the marginal utility of money is constant). To illustrate why Marshall’s theory needs this assumption, let’s go back to the T-shirt example, and assume that the consumer doesn’t face a single price for T-shirts. Instead, the first T-shirt always costs £15, and only after they have bought it, can they get the second one for £12. At first instance, one may conclude that the person would still buy the two T-shirts. After all, as it was argued before, the consumer values the first one at £15, and the second one at £12. But in the real world, this needn’t be the case. In this second scenario, a person needs to spend more money to buy the first T-shirt (£15 instead of £12), which could make them choose against buying the second one for £12. In other words, the only reason the consumer would buy the second T-shirt with a common price of £12, was because the first one cost £12 as well. This example suggests that the consumer could actually value the second T-shirt less than £12, as they may choose not to buy it at this price. The implication of the above scenario is that in order for Marshall’s consumer surplus to be meaningful, one needs to assume that a consumer’s wealth doesn’t affect their purchasing decisions (constant marginal utility of money assumption). In such a world, a consumer wouldn’t ever feel ‘poorer’, and they would always buy the second T-shirt for £12, even if the first one cost £15. While Marshall defended his assumption on the grounds that his theory should only be used to analyse small purchases, for which one’s wealth shouldn’t matter much, it still attracted severe criticism at his time (see Dooley, 1983). And even if one were to agree that his theory holds for small purchases, this still limits tremendously the scope for applying his theory in real-world scenarios. A breakthrough for Economics Although Marshall’s theory of consumer surplus has many issues, several of which aren’t discussed in this post, it has been clearly very influential on the economics profession, as is evident from its popularity in Economics textbooks. What’s more, it has provided a fertile ground for later theories of consumer surplus, such as the Hicksian consumer surplus, which doesn’t assume a constant marginal utility of money.

Bramley and Fitzpatrick (2018) investigate the impact of different factors in driving people to homelessness in the UK, and they refute the hypothesis that the event of homelessness is randomly distributed across the population. Childhood poverty, in particular, is found to be the best predictor of whether one experienced homelessness in the UK. Picture by Matt Collamer, Unsplash A ‘critical realist’ approach to homelessness The authors start by framing the event of homelessness as the result of a complex interplay between ‘individual’ (e.g. mental ill health, substance addictions) and ‘structural’ (e.g. poverty, housing market conditions) factors. Poverty and mental health illnesses, for example, can exacerbate each other through a feedback loop, which, in turn, can push one to homelessness. That said, they highlight how the relative importance of different factors can vary between places, depending on things such as local institutions. Countries with more generous social security policies, for example, may have a lower overall prevalence of homelessness, but among this homeless population, a higher proportion of the people may face complex personal problems (‘individual’ factors). Milburn et al’s (2007) cross-national study, for example, found that in Australia (with its relatively strong welfare regime) young homeless people faced more 'individual' difficulties than their peers in the US (with its much narrower welfare safety net). Methodology Bramley and Fitzpatrick (2018) use three different data sets (Scottish Household Survey, UK Poverty and Social Exclusion Survey, British Cohort Study 1970), which they analyze separately, to try and identify the key drivers of homelessness in the UK. The British Cohort Study (2000) dataset appears to be the most interesting one. Although it’s out-of-date, it’s the only panel dataset used, and it provides systematic data for every individual born in the UK in one specific week in 1970. This allows for a robust investigation of the effect of early-life factors, such as childhood poverty and teenage experiences, on the event of homelessness. Using this dataset, the authors estimate the impact of the following blocks of factors on the probability of one becoming homeless in the UK by the age of 30: Demographics, which refers to ‘fixed’ individual attributes such as gender and ethnicity. Childhood poverty, which is represented by three variables, two measured at age 10 (living in rented housing, lack of consumer durables) and one at age 16 (household income per head). Geography, which captures factors such as whether the person lived in a rural or urban setting. Teenage experiences, such as whether a person lived with both natural parents at age 16, and the state of the mother’s mental health. Adult economic situation, which includes educational, labour, and housing market factors experienced up to the age of 26. Adult family and life events, which has variables such as whether a person has a partner or children, or any long-term illness or disability. Results The results refute the hypothesis that the chances of becoming homeless are equally distributed across the population, at least back when the data was collected (1970-2000). This becomes apparent once we compare the predicted probabilities of experiencing homelessness for the following hypothetical people. First, take a white male who had a relatively affluent childhood in the rural south of England, an unproblematic school career, went to university and graduated at 21, who was living with his parents at age 26, with no partner relationship and no children. He has a predicted probability of experiencing homelessness by the age of 30 of just 0.6%. In contrast, take a mixed-ethnicity female, who experienced poverty as a child, was brought up by a lone parent, left school or college at 16, had spells of unemployment, and was living as a renter with no partner but with her own children at age 26. Her predicted probability is 71.2%. Meanwhile, it is striking that childhood poverty was found to be by far the best predictor of whether one would experience homelessness by the age of 30 (see Figure 1). This suggests that early-life policy interventions could be a powerful tool in preventing homelessness later on in life. What’s more, it provides evidence that ‘structural’ factors, rather than ‘individual’ ones, are stronger drivers of homelessness in the UK, at least if we are to look back to the 1980s and 1990s. Figure 1: Childhood poverty explained 52% of the occurrence of homelessness Data: Bramley and Fitzpatrick (2018), Bonsai Economics.