machine learning, Meta Intelligence

Science-Watching: From Ignition to Energy

[from Science & Technology Review July/August 2025 Research Highlights, by Noah Pflueger-Peters]

Achieving ignition at the National Ignition Facility (NIF) proved that harnessing the power of the Sun in a laboratory may be possible. The Sun’s extreme temperatures and pressures cause light elements to fuse together to create heavier ones, releasing enormous energy and sustaining conditions for more thermonuclear reactions. NIF replicates these conditions with inertial confinement fusion, in which lasers compress and heat a target capsule filled with deuterium and tritium (DT), “heavy” isotopes of hydrogen that contain extra neutrons. When the isotopes fuse, they create helium and a neutron, and the lost mass is converted into inertial fusion energy (IFE), which can be harnessed for energy production.

Nuclear fusion produces significantly more energy than either nuclear fission or burning fossil fuels for equivalent amounts of fuel. Since the input materials for fusion energy are plentiful on Earth, an IFE power plant could produce safe, abundant, power grid-compatible energy without highly radioactive byproducts.

Although significant work remains to harness fusion energy, pursuing the development and deployment of IFE is crucial for the nation’s energy security, enabling the United States to shape implementation worldwide, avoid technological surprises from adversaries, and influence technical leadership in other energy-intensive technologies such as AI, machine learning (ML), and supercomputing.

IFE research stretches back to the early days of Lawrence Livermore, and today the Laboratory is fostering the overall fusion ecosystem. Livermore’s unique capabilities, expertise, and connections will be critical to laying the technical, logistical, and legal groundwork to make IFE possible. “IFE is a grand scientific and engineering challenge, something that is so incredibly difficult and high-risk and takes enormous expertise,” says Tammy Ma, Livermore’s IFE Institutional Initiative lead. “This challenge makes it the right kind of problem for national laboratories to pursue.”

This artist’s rendering shows the concept for an inertial fusion energy (IFE) power plant design, with a cutaway to show the plant’s target chamber in the center. Livermore researchers are laying the groundwork for private fusion companies to build similar designs. (Illustration by Eric Smith.)

Designing for Viability

NIF is the only facility to date to demonstrate the ignition and burning plasma conditions that are prerequisites for IFE, but it is an experimental facility for stockpile stewardship research, not a power plant. To be commercially viable and produce the energy to offset costs and meet demands (baseload power), IFE plants will need to generate more than 30 times the energy they deliver to the fusion target on every shot while firing 10 or more shots per second, compared to NIF’s rate of one or two shots per day.

The Laser Inertial Fusion Energy (LIFE) study, conducted between 2008 and 2013, aimed to build directly on technology developed for NIF to achieve IFE and took a systematic approach to this requirement by developing the Integrated Process Model (IPM). (See S&TR, April/May 2009 [archived PDF], pp. 6-15.)

IPM is a technoeconomic model of an IFE power plant with detailed technical and cost breakdowns and interdependencies of key systems and subsystems. “The work done under LIFE was fantastic,” says Ma. “IPM lays out engineering and physics requirements for the entire system to test out different scenarios and see the impact. Now, we not only get to expand on all that but also leverage 15 years of new data from NIF, better codes, and high-performance computing (HPC), as well as new work in AI, ML, advanced manufacturing, diagnostics, and nonproliferation across the Laboratory.”

IPM describes an IFE power plant that requires a solid-state laser driver system to “pump” lasers with optical energy using laser diodes instead of flashlamps as at NIF. The plant will also need to fabricate and fill target capsules onsite and send them into its target chamber at a high enough frequency to produce baseload power. “We will have to repeatedly inject targets into the chamber, so the targets must be able to withstand and survive that process,” explains Ma. “Then, the lasers will track the moving targets, and when one gets to the center of the chamber, they would fire on the centered target, repeating 10 to 20 times per second.”

The facility would convert fusion energy into heat and then electricity via steam turbines, sending most of the electricity to the power grid and recycling the rest to power operations on subsequent shots. Neutrons from the reaction would produce tritium needed for the DT fuel by bombarding lithium isotopes in a “breeding blanket” material lining its target chamber. By closing both the power and fuel cycles, IFE plants are expected to be self-sustaining.

Thanks in part to IFE STARFIRE (IFE Science and Technology Accelerated Research for Fusion Innovation and Reactor Engineering), a Department of Energy (DOE)-funded multi-institutional IFE research and development hub, researchers across the Laboratory are working to meet the new system’s demands. IPM can help identify key challenges, test the viability of new designs, and direct future research. “Many technical models and cost models exist for IFE, but very few, if any, pair systems and cost models together at the same depth as IPM,” says Mackenzie Nelson, a technoeconomic systems analyst in the Computational Engineering Division. “This type of tool offers such an advantage because we can assess design choices from both a technical and economic standpoint and create blueprints for what an IFE plant could look like.”

(left to right) Livermore researchers Bassem El Dasher, Claudio Santiago, and Mackenzie Nelson discuss a 3D model of a proposed IFE power plant design alongside the Integrated Process Model (IPM). IPM has more than 270 potential user inputs that researchers and collaborators can use to assess different IFE design choices to see the technical and cost impact on the entire design.

Operational Demands

NIF’s target capsules are extremely precise, fragile, and can take weeks to fabricate, fill, and position. Researchers are trying to reconcile that factor with the estimated demand of more than 800,000 capsules per day produced at less than $0.50 each to achieve IFE plant viability. To do this, they are examining optimal target designs for IFE and exploring advanced manufacturing methods such as microfluidics, volumetric additive manufacturing, and two-photon polymerization. (See S&TR, April/May 2025 [archived PDF], pp. 16-19.) Additional projects involve developing diagnostic instruments that can collect, analyze, and combine data with other diagnostics at the 10 to 20 shot per second frequency and use it to improve lasers in real time.

Fusion energy systems such as IFE are also a regulatory challenge, as they generate high-energy neutrons capable of breeding plutonium or uranium-233 and rely on large quantities of tritium. “Pure fusion energy systems do not require fissile material, but there are still ways to misuse these technologies that pose proliferation risk,” says Yana Feldman, the associate program leader for international safeguards. Bad actors may only need small amounts of tritium to make nuclear weapons, and some breeding blanket designs may inadvertently produce traces of plutonium that may be diverted for military purposes.

Nuclear fission reactors are regulated through international agreements and export control rules, and the independent International Atomic Energy Agency (IAEA) verifies that nuclear material and facilities are only being used for peaceful purposes. Neither treaties nor the IAEA address fusion energy, and no consensus has been reached on whether fusion energy systems need an international verification program. Verification methods for safeguarding tritium are also far less developed than for plutonium and uranium and focus more on contamination and transfers than analytical accounting for discrepancies. The precise scale of allowable tritium unaccounted for without posing proliferation risk is also unclear.

Fusion systems can be designed for proliferation resistance, but not having an existing design remains a challenge.

International security analyst Anne-Marie Riitsaar and her colleagues are exploring these complexities and starting conversations with international fusion experts and private industry to raise awareness. Riitsaar also plans to collaborate with the IPM team to map tritium diversion vulnerabilities and identify high-risk points where researchers could incorporate surveillance methods into plant designs to detect and prevent potential misuse. “People sometimes ask me why I’m thinking about fusion energy regulations and proliferation risks at this point, but it’s not too early,” says Riitsaar. “Reaching a multinational consensus on regulating sensitive technologies takes considerable time and effort.”

The National Ignition Facility is an experimental facility and not a power plant, so a commercial IFE plant design has vastly different requirements—many of which are being studied by Livermore researchers and their collaborators.

	NIF	Viable IFE plant (estimated)
Repetition rate	One shot per day	10 to 20 shots per second
Energy gain	4.13 times (as of April 2025)	30 times (minimum), 50 times to 100 times (ideal)
How lasers gain energy	Flashlamps	Diode pumping
Target fabrication and fuel filling	Fabricated offsite over several weeks and filled manually in 1 to 5 days	Mass-manufactured and filled in a target factory within the facility
Target delivery	Positioned manually within the Target Chamber	Shot into the plant’s target chamber approximately 10 to 20 times per second
Laser alignment	Computationally in real time, taking up to 8 hours	In real time
Power cycle	Open, requiring outside energy sources	Closed, applying reused energy to power laser and ancillary plant operations
Fuel cycle (tritium)	Produced offsite	Bred onsite

The Laser Driven Fusion Integration Research and Science Test Facility (LD-FIRST) is a proposed blueprint for a proof-of-concept IFE facility that will test all the key IFE subsystems in an integrated fashion. A public-private partnership will likely be necessary to build the facility and will help the IFE community address the main subset of risks and the technological challenges of building a commercial plant.

Converging on a Solution

The team seeks to make IPM as accurate and comprehensive as possible by meeting with subject matter experts across the Laboratory to incorporate the latest research. “We’re trying to evolve the model so it has the same level of high detail across every single functional area to tell us where we can focus research and help us find optimized solutions that we could propose to industry,” says Nelson.

Computer scientist Claudio Santiago and his colleagues also modernized IPM by porting its framework from Microsoft Excel to Python in December 2024, making it compatible with AI, ML, design optimization, and HPC to further inform designs. “Once we think about all the forcing functions such as minimum shot yield and materials requirements pinning us in from every direction, we end up with an optimized solution space. As we sharpen the pencil more with these tools, that optimized solution box gets smaller until eventually we’ve converged on a point design,” says IFE lead systems engineer Justin Galbraith. Galbraith and his team’s point design is called the Laser Driven Fusion Integration Research and Science Test Facility, or LD-FIRST, a proof-of-concept physics demonstration facility for IFE. “That point design, we anticipate, will serve as the foundation for a future public-private partnership that would facilitate building and realizing a physical facility to focus the IFE community in pursuit of fusion power on the grid,” says Galbraith.

Livermore is leading the charge in IFE, helping the United States develop a technological roadmap, growing and coordinating science and technology efforts within the Laboratory, and fostering partnerships across the fusion industry, academia, and government.

Ma chaired DOE’s “Basic Research Needs for IFE” workshop and report in 2022 and co-chairs the subcommittee providing recommendations on the nation’s fusion activities through DOE’s Fusion Energy Sciences Advisory Committee. She and her team travel often to Washington, D.C., working with DOE and legislators to expand fusion energy research and advocacy in the nation. Livermore also leads a “Collaboratory” with other DOE national laboratories to connect research project leads and facilitate public-private partnerships. The Collaboratory has hosted multiple events with industry, and the Laboratory has partnered with three private companies who aim to design pilot IFE plants.

Meanwhile, Galbraith and other IFE leaders have served as technical advisors for engineering design teams at Texas A&M University and given them IFE-relevant problems to solve, including advanced chamber and blanket design. Galbraith is working with Nelson to develop the IFE plant design portion of a high-energy-density science summer school program, which Nelson is leading in 2025 at the University of California at San Diego, and they have developed IFE curriculum that has been deployed at six universities starting in spring 2025. “We’re hoping we can get a group of students really excited about fusion and start to build up the next generation of engineers and scientists that will make fusion a reality,” says Galbraith. The team has led IFE strategic planning exercises at the Laboratory, and Lawrence Livermore will stand up a new fusion institute—named “LIFT,” for Livermore Institute for Fusion Technology—a research and development center that will coordinate and centralize institutional fusion energy research.

Harnessing IFE will be a massive undertaking, but Livermore’s broad and deep expertise, facilities, and capabilities put the Laboratory in a unique position to lead and play an impactful role. “If we can set it up correctly, IFE will be a big piece of the Laboratory’s long-term vision,” says Ma. “IFE plays off of our history and all of our strengths, and it is critical for long-term national security.”

Science-Watching: Forecasting New Diseases in Low-Data Settings Using Transfer Learning

[from London Mathematical Laboratory]

by Kirstin Roster, Colm Connaughton & Francisco A. Rodrigues

Abstract

Recent infectious disease outbreaks, such as the COVID-19 pandemic and the Zika epidemic in Brazil, have demonstrated both the importance and difficulty of accurately forecasting novel infectious diseases. When new diseases first emerge, we have little knowledge of the transmission process, the level and duration of immunity to reinfection, or other parameters required to build realistic epidemiological models. Time series forecasts and machine learning, while less reliant on assumptions about the disease, require large amounts of data that are also not available in early stages of an outbreak. In this study, we examine how knowledge of related diseases can help make predictions of new diseases in data-scarce environments using transfer learning. We implement both an empirical and a synthetic approach. Using data from Brazil, we compare how well different machine learning models transfer knowledge between two different dataset pairs: case counts of (i) dengue and Zika, and (ii) influenza and COVID-19. In the synthetic analysis, we generate data with an SIR model using different transmission and recovery rates, and then compare the effectiveness of different transfer learning methods. We find that transfer learning offers the potential to improve predictions, even beyond a model based on data from the target disease, though the appropriate source disease must be chosen carefully. While imperfect, these models offer an additional input for decision makers for pandemic response.

Introduction

Epidemic models can be divided into two broad categories: data-driven models aim to fit an epidemic curve to past data in order to make predictions about the future; mechanistic models simulate scenarios based on different underlying assumptions, such as varying contact rates or vaccine effectiveness. Both model types aid in the public health response: forecasts serve as an early warning system of an outbreak in the near future, while mechanistic models help us better understand the causes of spread and potential remedial interventions to prevent further infections. Many different data-driven and mechanistic models were proposed during the early stages of the COVID-19 pandemic and informed decision-making with varying levels of success. This range of predictive performance underscores both the difficulty and importance of epidemic forecasting, especially early in an outbreak. Yet the COVID-19 pandemic also led to unprecedented levels of data-sharing and collaboration across disciplines, so that several novel approaches to epidemic forecasting continue to be explored, including models that incorporate machine learning and real-time big data data streams. In addition to the COVID-19 pandemic, recent infectious disease outbreaks include Zika virus in Brazil in 2015, Ebola virus in West Africa in 2014–16, Middle East respiratory syndrome (MERS) in 2012, and coronavirus associated with severe acute respiratory syndrome (SARS-CoV) in 2003. This trajectory suggests that further improvements to epidemic forecasting will be important for global public health. Exploring the value of new methodologies can help broaden the modeler’s toolkit to prepare for the next outbreak. In this study, we consider the role of transfer learning for pandemic response.

Transfer learning refers to a collection of techniques that apply knowledge from one prediction problem to solve another, often using machine learning and with many recent applications in domains such as computer vision and natural language processing. Transfer learning leverages a model trained to execute a particular task in a particular domain, in order to perform a different task or extrapolate to a different domain. This allows the model to learn the new task with less data than would normally be required, and is therefore well-suited to data-scarce prediction problems. The underlying idea is that skills developed in one task, for example the features that are relevant to recognize human faces in images, may be useful in other situations, such as classification of emotions from facial expressions. Similarly, there may be shared features in the patterns of observed cases among similar diseases.

The value of transfer learning for the study of infectious diseases is relatively under-explored. The majority of existing studies on diseases remain in the domain of computer vision and leverage pre-trained neural networks to make diagnoses from medical images, such as retinal diseases, dental diseases, or COVID-19. Coelho and colleagues (2020) explore the potential of transfer learning for disease forecasts. They train a Long Short-Term Memory (LSTM) neural network on dengue fever time series and make forecasts directly for two other mosquito-borne diseases, Zika and Chikungunya, in two Brazilian cities. Even without any data on the two target diseases, their model achieves high prediction accuracy four weeks ahead. Gautam (2021) uses COVID-19 data from Italy and the USA to build an LSTM transfer model that predicts COVID-19 cases in countries that experienced a later pandemic onset.

These studies provide empirical evidence that transfer learning may be a valuable tool for epidemic forecasting in low-data situations, though research is still limited. In this study, we aim to contribute to this empirical literature not only by comparing different types of knowledge transfer and forecasting algorithms, but also by considering two different pairs of endemic and novel diseases observed in Brazilian cities, specifically (i) dengue and Zika, and (ii) influenza and COVID-19. With an additional analysis on simulated time series, we hope to provide theoretical guidance on the selection of appropriate disease pairs, by better understanding how different characteristics of the source and target diseases affect the viability of transfer learning.

Zika and COVID-19 are two recent examples of novel emerging diseases. Brazil experienced a Zika epidemic in 2015–16 and the WHO declared a public health emergency of global concern in February 2016. Zika is caused by an arbovirus spread primarily by mosquitoes, though other transmission methods, including congenital and sexual have also been observed. Zika belongs to the family of viral hemorrhagic fevers and symptoms of infection share some commonalities with other mosquito-borne arboviruses, such as yellow fever, dengue fever, or chikungunya. Illness tends to be asymptomatic or mild but can lead to complications, including microcephaly and other brain defects in the case of congenital transmission.

Given the similarity of the pathogen and primary transmission route, dengue fever is an appropriate choice of source disease for Zika forecasting. Not only does the shared mosquito vector result in similar seasonal patterns of annual outbreaks, but consistent, geographically and temporally granular data on dengue cases is available publicly via the open data initiative of the Brazilian government.

COVID-19 is an acute respiratory infection caused by the novel coronavirus SARS-CoV-2, which was first detected in Wuhan, China, in 2019. It is transmitted directly between humans via airborne respiratory droplets and particles. Symptoms range from mild to severe and may affect the respiratory tract and central nervous system. Several variants of the virus have emerged, which differ in their severity, transmissibility, and level of immune evasion.

Influenza is also a contagious respiratory disease that is spread primarily via respiratory droplets. Infection with the influenza virus also follows patterns of human contact and seasonality. There are two types of influenza (A and B) and new strains of each type emerge regularly. Given the similarity in transmission routes and to a lesser extent in clinical manifestations, influenza is chosen as the source disease for knowledge transfer to model COVID-19.

For each of these disease pairs, we collect time series data from Brazilian cities. Data on the target disease from half the cities is retained for testing. To ensure comparability, the test set is the same for all models. Using this empirical data, as well as the simulated time series, we implement the following transfer models to make predictions.

Random forest: First, we implement a random forest model which was recently found to capture well the time series characteristics of dengue in Brazil. We use this model to make predictions for Zika without re-training. We also train a random forest model on influenza data to make predictions for COVID-19. This is a direct transfer method, where models are trained only on data from the source disease.
Random forest with TrAdaBoost: We then incorporate data from the target disease (i.e., Zika and COVID-19) using the TrAdaBoost algorithm together with the random forest model. This is an instance-based transfer learning method, which selects relevant examples from the source disease to improve predictions on the target disease.
Neural network: The second machine learning algorithm we deploy is a feed-forward neural network, which is first trained on data of the endemic disease (dengue/influenza) and applied directly to forecast the new disease.
Neural network with re-training and fine-tuning: We then retrain only the last layer of the neural network using data from the new disease and make predictions on the test set. Finally, we fine-tune all the layers’ parameters using a small learning rate and low number of epochs. These models are examples of parameter-based transfer methods, since they leverage the weights generated by the source disease model to accelerate and improve learning in the target disease model.
Aspirational baseline: We compare these transfer methods to a model trained only on the target disease (Zika/COVID-19) without any data on the source disease. Specifically, we use half the cities in the target dataset for training and the other half for testing. This gives a benchmark of the performance in a large-data scenario, which would occur after a longer period of disease surveillance.

The remainder of this paper is organized as follows. The models are described in more technical detail in Section 2. Section 3 shows the results of the synthetic and empirical predictions. Finally, Section 4 discusses practical implications of the analyses.

Access the full paper [via institutional access or paid download].

Economics—Atlanta Fed’s November Notes from the Vault

Financial System of the Future

Notes from the Vault
Larry D. Wall
November 2019

The financial services business is an information technology business that operates in an environment with substantial government involvement. The state of both technology and government intervention have evolved considerably since the financial crisis that began in 2007, with potentially large implications for the future financial system. With that in mind, the Atlanta Fed recently hosted the workshop Financial System of the Future, which was cosponsored by the Center for the Economic Analysis of Risk (CEAR) at Georgia State University.

This post reviews many of the papers and presentations from the workshop dealing with both financial regulation and innovations in financial technology. It also reviews one of the keynote speeches on the important topic of corporate governance. A companion macroblog post summarizes the workshop’s discussion of digital currencies.

Post-crisis changes in regulation

Some banks had difficulty obtaining adequate funding from private financial markets during the crisis of 2007–08, despite being well capitalized by that period’s capital standards. In part, these funding difficulties were a result of bank depositors fearing that banks had not adequately provisioned for their credit losses.

A paper by University of Michigan professors Thomas Flanagan and Amiyatosh Purnanandam examines the question “Why Do Banks Hide Losses?” [PDF] with a particular focus on the role of shareholder monitoring and management incentives. One problem in conducting such a study is that of separating bad investment decisions from deceptive accounting. This paper exploits an unexpected change in regulation in India to help provide such separation. This regulatory change forced all of the banks in that country to detail the extent of their underreporting of loan losses in 2015. The paper finds that weaker shareholder monitoring and higher-power executive compensation contracts are associated with more underreporting of loan losses.¹ This paper helps us to understand how banks exploit accounting discretion over loan losses and reinforces the importance of supervisory oversight.

The funding problems encountered by U.S. banks led to runs and ultimately the collapse of a few banks, but these problems had a significant effect on the operations of some banks that survived. One measure of the vulnerability of a bank to a funding shock was its reliance on uninsured, wholesale deposits from large institutions. Professors Sudheer Chava of Georgia Institute of Technology, Rohan Ganduri of Emory University, Linghang Zeng of Babson College, and graduate student Nikhil Paradkar of Georgia Tech examine one effect of these shocks in their paper “Shocked by Bank Funding Shocks: Evidence from 500 Million Consumer Credit Cards” [PDF]. The paper finds that banks that were more exposed to wholesale funding shocks made larger cuts in consumer credit lines than other banks. The authors further find that these cutbacks were made to the most credit-constrained customers (those with lower credit scores and higher utilization of their credit line). The results also suggest that these credit-constrained consumers responded by cutting back their consumption. Overall, these results provide additional support for regulatory policies intended to strengthen banks so that they can continue lending through future crises.

The Federal Reserve was created in 1913 in large part to provide a source of liquidity during periods of stress in bank lending markets. Yet banks proved surprisingly resistant to borrowing from the Fed’s discount window at the start of the funding problems in the fall of 2007. In an attempt to overcome this reluctance to borrow, the Fed created the Term Auction Facility (TAF) in December 2007. A paper by professors Yunzhi Hu of the University of North Carolina at Chapel Hill and Hanzhe Zhang from Michigan State University examines the TAF in a paper titled “Overcoming Borrowing Stigma: The Design of Lending-of-Last-Resort Policies” [PDF]. This paper develops a simple model in which TAF funding is preferred by stronger banks because the funds are only available with a delay (that is, the funds will not help a bank facing an immediate run) and because TAF gives stronger banks the potential to borrow at lower costs. The paper also provides evidence that when both the discount window and TAF were available, the weakest banks borrowed from the discount window and the relatively stronger banks bid for TAF funds. The results of this study may help to inform the Fed’s policies on the provision of emergency liquidity in the future.

In the post-crisis period, the bank supervisors agreed to adopt new liquidity rules and strengthen existing capital requirements at the Basel Committee on Banking Supervision. A paper by Harvard University professor Daniel Roberts and Federal Reserve Bank of New York economists Asani Sarkar and Or Shachar examines the Basel liquidity requirements in the paper “Bank Liquidity Creation, Systemic Risk, and Basel Liquidity Regulations” [PDF]. The paper finds that the new regulation has the cost of reducing bank liquidity creation in normal times but also results in more resilient banks in a crisis. The authors estimate both the costs and benefits of the regulation and find that the benefits in terms of a more stable financial system exceed the costs of the reduction in liquidity.

Josef Schroth, an economist at the Bank of Canada, analyzes the question of how regulators could use capital regulation to smooth bank lending through a recession and financial crisis in his paper “Macroprudential Policy with Capital Buffers” [PDF]. Banks in his model hold sufficient capital in most periods but have insufficient capital to retain uninsured market funding during crises, which leads to severe credit crunches. Bank regulators can smooth lending by requiring banks to hold more capital during good times, also allowing banks to operate with reduced capital during a crisis and rebuild their capital gradually after a crisis. The paper’s findings are consistent with the Basel III’s creation of a capital conservation buffer and countercyclical capital buffer.

Issues related to bank capital have been the subject of ongoing discussion among academics, bankers, and regulators since the crisis. Bank of Finland economists Gene Ambrocio and Esa Jokivuolle, along with professors Iftekhar Hasan from Fordham University and Kim Ristolainen from the University of Turku, surveyed the views of banking and finance academics and reported the results in the paper “Are Bank Capital Requirements Optimally Set? Evidence from Researchers’ Views” [PDF] (slide deck [PDF]). The survey obtained results from 149 leading academic researchers. It found that most respondents believe that higher capital ratios would reduce the likelihood of a crisis and the social costs associated with a crisis. Most respondents thought an increase of 5 percentage points in capital would increase the average cost of capital and reduce bank lending. However, a majority said that this reduction in lending would result in “minimal to no change” in economic activity.

Fintech lending

Banks have historically been important suppliers of credit to consumers and small businesses. However, since the crisis, lending by some fintech firms to these borrowers has grown rapidly. The conference addressed a variety of questions such as the sources of competitive advantage to fintech lenders; competition between fintech lenders, small banks, and large banks; and the impact of fintech lending on consumers. Additionally, the conference looked at post-crisis changes in bank securitization rules, which may have some relevance to fintech lenders’ efforts to fund their loans.

Banks, especially large banks, have long used statistical analysis to help them predict the credit riskiness of consumers and small businesses. However, fintech firms have promoted their use of machine learning techniques as resulting in better credit analysis. Economist J. Christina Wang and former research associate Charles B. Perkins from the Federal Reserve Bank of Boston explore the value added of some machine learning techniques in their paper “How Magic a Bullet Is Machine Learning for Credit Analysis? An Exploration with Fintech Lending Data” [PDF]. In particular, they examine the contribution of tree-based models using data from LendingClub, a fintech lender. Tree-based models provide easily understood decision rules that are valuable for demonstrating compliance with U.S. fair lending rules. The results indicate that tree-based machine learning improves prediction accuracy, in part by uncovering notable interactive effects. These models benefit from additional data (albeit only up to about 5,000 observations) and from the addition of more predictors. However, exactly which tree-based approach worked best varied across different subgroups of consumers.

Many large banks retreated from some types of retail and small business lending. In part, this pullback was likely to due to a combination of the increased riskiness of these loans and the tightening of bank capital requirements. However, fintech lenders have been expanding into some of the same spaces, raising the question of whether the reduced lending by big banks was due in part to competitive pressure from fintech firms. It also raises the question of whether smaller banks were also retreating due to increased competition or expanding to help fill the void left by the larger banks. Two papers examine the rise of fintechs in two different lending markets and analyze what developments in these markets tell us about big versus small banks.

Professors Taylor Begley from Washington University in St. Louis and Kandarp Srinivasan from Northeastern University focus on mortgage market lending in their paper “Small Bank Lending in the Era of Fintech and Shadow Banking: A Sideshow?” [PDF]. Their paper observes that fintech lenders rely largely on government-supported loan securitization programs such as those of Fannie Mae, Freddie Mac, and the Federal Housing Administration. However, not all mortgage loans qualify for these programs in large part due to limits on the maximum size of the loan. Thus, fintech mortgage lenders were able to take nationwide market share in government-supported loans from the large banks. Nevertheless, it was the small banks that stepped in to provide mortgage loans that did not qualify for government-supported securitization. As a result, small banks have proven more responsive to the withdrawal of bigger banks and have maintained nearly constant market share through the period from 2009 to 2013.

The paper “What Is Fueling the Fintech Lending Revolution? Local Banking Market Structure and Fintech Market Penetration” [PDF] by professor Tetyana Balyuk of Emory University and professors Allen N. Berger and John Hackney from the University of South Carolina studies the rise of fintech lenders in small business lending. The authors find that fintech lenders’ market penetration is associated with lower small bank market shares, consistent with the decline of small banks over time and helping to explain fintech lenders’ success. They also find evidence that small banks tend to make safer loans than fintech firms. Finally, looking at the contribution of different lenders to employment, the authors find that fintech matters most for the smallest businesses, small banks matter more for larger small businesses, and neither matters for large businesses.

As fintech lending has grown, the question of whether it is good for consumers has also arisen. This may be a trick question. We have long known that if people are given a choice, some will use it wisely and others will not. Last year’s CEAR–CenFIS conference provided some evidence of how consumers manage their finances after borrowing from fintech lenders (see “Winners and Losers of Marketplace Lending: Evidence From Borrower Credit Dynamics” [PDF] and “The Real Effects of Financial Technology: Marketplace Lending and Personal Bankruptcy” [PDF]). This year, Federal Reserve Board economists Traci Mach and Timothy Dore provided additional evidence in the paper “Marketplace Lending and Consumer Credit Outcomes: Evidence from Prosper” [PDF]. The authors were able to match application data from one lender to credit bureau data. They find that relative to applicants without an origination, borrowers see an increase in their credit scores and total debt, but decreases in their delinquency rates. The increase in mortgage debt balances was especially notable.

An important issue for any type of lender is its access to funding. Banks fund their loans with a combination: (a) their own liabilities in the form of deposits and other market borrowings and (b) loan sales to investors. The risk with on-balance-sheet lending is that too many loans will turn bad, the bank will suffer large loan losses, and depositors will flee the bank. Although some fintech firms fund loans with their own liabilities, most of their loans are funded by sales to investors. The advantage to the loan originator of selling loans is that, if the loan goes bad, the buyers of the loan bear the credit losses. However, such sales can lead to concerns that the bank or fintech originating a loan will not be as careful in underwriting and monitoring as they would be with the loans they retain on their own books. The result can be similar to that discussed in the Chava et al. paper, where originators have to reduce their lending volume.

One alternative that may have the effect of improving banks’ incentives and thereby mitigating investor concerns is that of having the bank retain part of the credit risk on the loans it sells. Professor Martin Hibbeln and graduate student Werner Osterkamp from the University of Duisburg-Essen study the effect of bank loan retention in their paper “The Impact of Skin in the Game on Bank Behavior in the Securitization Market” [PDF]. The authors exploit changes in European Union regulations to study loans sold by European banks in the securitization market. They find evidence that retention of part of the risk of a loan does not improve the quality of the loans being securitized.

Corporate governance

The role of shareholders in corporate governance is important for both financial and non-financial firms. The last few decades have seen shareholder governance power increasingly concentrated in institutional investors such as mutual funds and pension funds. Although a few of these institutional funds are explicitly focused on corporate governance issues, most are not. However, even the funds that are not focused on governance nevertheless have an interest in participating in governance decisions. The problem for these investors is how to keep up with the many issues requiring a shareholder vote. The answer for a large fraction of institutional investors is to seek information and advice from proxy advisory firms. The role of these advisory firms was the topic of Carnegie Mellon University professor Chester Spatt’s dinner speech titled “Proxy Advisory Firms, Governance, Market Failure, and Regulation” [PDF] (See also his related paper [PDF].)

Spatt observed that the importance of proxy advisory firms is easy to understand. All shareholders in a firm face the same questions, and there is similarity in questions across firms. However, any one shareholder obtains only a small fraction of the gains from better governance while potentially paying all of the costs of analysis. In this setting, proxy advisory firms are a predictable market solution to a potential free-riding problem. The increasing returns to scale that arise in this setting also naturally lends itself to the development of a monopoly—or in this case, a duopoly. The result can be reduced incentives for the two proxy advisory firms to avoid factual mistakes and increased power for the advisers to impose their philosophy on firms. Spatt discussed a variety of steps the Securities and Exchange Commission has taken and further actions that could mitigate these concerns. However, the problem of what to do about the role of institutions and proxy advisers in corporate governance is one that may not be totally solvable, but rather requires careful management over time.

Conclusion

The post-crisis period has seen dramatic changes in both information technology and government regulation. The papers presented recently at a workshop at the Atlanta Fed analyzed a number of these changes and contribute to our understanding of how the financial system is likely to evolve in the future.

Larry D. Wall is executive director of the Center for Financial Innovation and Stability at the Atlanta Fed. The author thanks Brian Robertson for helpful comments. The view expressed here are the author’s and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. If you wish to comment on this post, please email atl.nftv.mailbox@atl.frb.org.

¹ For further discussion on banks’ reporting of loan losses, see Larry’s recent post on changes to U.S. accounting rules.

Physics AI Predicts That Earth Goes Around the Sun

from Nature Briefing:

Hello Nature readers,

Today we learn that a computer Copernicus has rediscovered that Earth orbits the Sun, ponder the size of the proton and see a scientific glassblower at work.

Physicists have designed artificial intelligence that thinks like the astronomer Nicolaus Copernicus by realizing the Sun must be at the center of the Solar System. (NASA/JPL/SPL)

AI ‘Discovers’ That Earth Orbits the Sun [PDF]

A neural network that teaches itself the laws of physics could help to solve some of physics’ deepest questions. But first it has to start with the basics, just like the rest of us. The algorithm has worked out that it should place the Sun at the centre of the Solar System, based on how movements of the Sun and Mars appear from Earth.

The machine-learning system differs from others because it’s not a black that spits out a result based on reasoning that’s almost impossible to unpick. Instead, researchers designed a kind of ‘lobotomized’ neural network that is split into two halves and joined by just a handful of connections. That forces the learning half to simplify its findings before handing them over to the half that makes and tests new predictions.

Next FDA Chief Will Face Ongoing Challenges

U.S. President Donald Trump has nominated radiation oncologist Stephen Hahn to lead the Food and Drug Administration (FDA). If the Senate confirms Hahn, who is the chief medical executive of the University of Texas MD Anderson Cancer Center, he’ll be leading the agency at the centre of a national debate over e-cigarettes, prompted by a mysterious vaping-related illness [archived PDF] that has made more than 2,000 people sick. A former FDA chief says Hahn’s biggest challenge will be navigating a regulatory agency under the Trump administration, which has pledged to roll back regulations.

Do We Know How Big a Proton Is? [PDF]

A long-awaited experimental result has found the proton to be about 5% smaller than the previously accepted value. The finding seems to spell the end of the ‘proton radius puzzle’: the measurements disagreed if you probed the proton with ordinary hydrogen, or with exotic hydrogen built out of muons instead of electrons. But solving the mystery will be bittersweet: some scientists had hoped the difference might have indicated exciting new physics behind how electrons and muons behave.

Contingency Plans for Research After Brexit

The United Kingdom should boost funding for basic research and create an equivalent of the prestigious European Research Council (ERC) if it doesn’t remain part of the European Union’s flagship Horizon Europe research-funding program [archived PDF]. That’s the conclusion of an independent review of how UK science could adapt and collaborate internationally after Brexit — now scheduled for January 31, 2020.

Nature’s 150th anniversary

A Century and a Half of Research and Discovery

This week is a special one for all of us at Nature: it’s 150 years since our first issue, published in November 1869. We’ve been working for well over a year on the delights of our anniversary issue, which you can explore in full online.

10 Extraordinary Nature Papers

A series of in-depth articles from specialists in the relevant fields assesses the importance and lasting impact of 10 key papers from Nature’s archive. Among them, the structure of DNA, the discovery of the hole in the ozone layer above Antarctica, our first meeting with Australopithecus and this year’s Nobel-winning work detecting an exoplanet around a Sun-like star.

A Network of Science

The multidisciplinary scope of Nature is revealed by an analysis of more than 88,000 papers Nature has published since 1900, and their co-citations in other articles. Take a journey through a 3D network of Nature’s archive in an interactive graphic. Or, let us fly you through it in this spectacular 5-minute video.

Then dig deeper into what scientists learnt from analyzing tens of millions of scientific articles for this project.

150 Years of Nature, in Graphics

An analysis of the Nature archive reveals the rise of multi-author papers, the boom in biochemistry and cell biology, and the ebb and flow of physical chemistry since the journal’s first issue in 1869. The evolution in science is mirrored in the top keywords used in titles and abstracts: they were ‘aurora’, ‘Sun’, ‘meteor’, ‘water’ and ‘Earth’ in the 1870s, and ‘cell’, ‘quantum’, ‘DNA’, ‘protein’ and ‘receptor’ in the 2010s.

Evidence in Pursuit of Truth

A century and a half has seen momentous changes in science, and Nature has changed along with it in many ways, says an Editorial in the anniversary edition. But in other respects, Nature now is just the same as it was at the start: it will continue in its mission to stand up for research, serve the global research community and communicate the results of science around the world.

Features & Opinion

Nature covers: from paste-up to Photoshop

Nature creative director Kelly Krause takes you on a tour of the archive to enjoy some of the journal’s most iconic covers, each of which speaks to how science itself has evolved. Plus, she touches on those that didn’t quite hit the mark, such as an occasion of “Photoshop malfeasance” that led to Dolly the sheep sporting the wrong leg.

Podcast: Nature bigwigs spill the tea

In this anniversary edition of Backchat, Nature editor-in-chief Magdalena Skipper, chief magazine editor Helen Pearson and editorial vice president Ritu Dhand take a look back at how the journal has evolved over 150 years, and discuss the part that Nature can play in today’s society. The panel also pick a few of their favorite research papers that Nature has published, and think about where science might be headed in the next 150 years.

Where I Work

Scientific glassblower Terri Adams uses fire and heavy machinery to hand-craft delicate scientific glass apparatus. “My workbench hosts an array of tools for working with glass, many of which were custom-made for specific jobs,” says Adams. “Each tool reminds me of what I first used it for and makes me consider how I might use it again.” (Leonora Saunders for Nature)

Quote of the Day

“At the very least … we should probably consider no longer naming *new* species after awful humans.”

Scientists should stop naming animals after terrible people — and consider renaming the ones that already are, argues marine conservation biologist and science writer David Shiffman. (Scientific American)

Yesterday was Marie Skłodowska Curie’s birthday, and for the occasion, digital colorist Marina Amaral breathed new life into a photo of Curie in her laboratory.

(If you have recommended people before and you want them to count, please ask them to email me with your details and I will make it happen!) Your feedback, as always, is very welcome at briefing@nature.com.

Flora Graham, senior editor, Nature Briefing