A Way Forward on Gun Safety: A Doable Reform That Could Make a Difference

This post was updated on October 4, with the addition of the Executive Summary and some minor editing of the text.  Several further changes were made on October 25 to clarify certain points, in response to comments received.  The substance has not been changed.  

Executive Summary

Deaths due to guns are far too high in the US – see the chart above.  And following the all-too-frequent mass shootings in the country, including the one at Uvalde, Texas, earlier this year, calls are made for something to be done.  Yet little is ever achieved.  While new gun safety legislation was passed following Uvalde, it was modest at best.  No one expects any measure that might put a significant dent in the number who die each year from firearms could ever be passed by the US Congress.  But a different approach is possible.

Two measures that have been proposed in the past, but which have not been put together, could have a far greater impact.  As importantly, the measures could be implemented through actions that a president has authority to take, coupled with actions that could be implemented in the states most willing to address gun safety.  As a track record is created and confidence is gained, it could then spread to further states.

The first pillar is to make available reliable personalized firearms (also often called “smart guns”), that will fire only when the owner is pulling the trigger and not when anyone else is.  The technology exists, but like any technology can be further improved.  One approach is fingerprint identification – a technology that has now been placed on hundreds of millions of smartphones worldwide, as well as on laptops, keyboards, and other devices.

Personalized weapons will not fire for others.  No longer could a child, when finding a parent’s loaded gun, end up tragically shooting themselves or a playmate.  Nor could a personalized firearm be used by a teen or preteen who, tragically, might be suicidal.  Nor would a personalized gun that had been stolen or obtained in some other way be of any use to teens in a gang – they would not fire (and one has to be 21 to buy a handgun).

Personalized firearms would also be safer for police and others.  Approximately 10% of police killed in the line of duty are killed with their own service weapon (or that of a colleague) – a weapon that was seized in a struggle with the officer as they tried to subdue a subject.  Similarly, private individuals confronting a criminal have often ended up being shot with their own gun.  (And while certainly not something I would recommend, there are those who believe teachers in schools should keep loaded and easily accessible guns in their classrooms.  A personalized firearm would not fire if a disgruntled student grabbed it.)

But perhaps the greatest value of personalized firearms that are linked to, and will fire only for, an individual owner is that they would change the dynamics of how guns end up in the hands of criminals. Stolen guns would be of no value to them – they could not be fired.  The same would be true for guns obtained through a straw purchaser (a colleague or a girlfriend), or from another gang member, or purchased on the black market.  Most criminals are armed with such illegally obtained weapons.  Personalized firearms would be of no use to them.

Making personalized firearms available would be complemented with an incentive to switch to them.  This is the second pillar, which would require gun owners to hold liability insurance that would pay compensation for any unjust harm caused by their gun.  This proposal comes from Jason Abaluck and Ian Ayres (professors at Yale).  The basic idea is similar to the requirement that all car owners hold insurance to cover the liability resulting from damages that car might cause – something required in all 50 states.

The payments would be at a standard amount (based on the nature of the injury) set by an insurance regulator.  This is similar to what is done for workers’ compensation insurance, and would allow the parties to avoid going to court to determine in each case the compensation to be paid.  Court processes are slow and expensive.

Private insurers would then have a strong incentive to determine (in competition with other insurers) what the insurance premium rates would need to be in order to cover the risks.  They would need to assess the risks and charge accordingly.  As for car insurance, those risks will likely be assessed based on factors such as the age of the owner, their gender, any criminal or other such record, the nature of the firearm being insured, and more.  Rates for hunting rifles, for example, would be relatively low, while those for handguns higher.  Importantly, the risks and hence the rates on personalized firearms would be well below what they would be for similar traditional weapons in similar hands.

Compensation payments in cases of suicides would need to be handled differently.  Just like for life insurance (which is not paid on suicides), paying the estates of those who commit suicide with their firearm could encourage such a tragic event.  Rather, the compensation payments from the insurer in such cases would be paid into a general fund.  That fund would be used for compensation payments to those harmed by a firearm, but where it was not possible to trace the origin of the firearm used.

The third pillar is that it can be implemented in a step-by-step process that does not require new federal legislation.  To start, the Biden administration would work with existing as well as potential new manufacturers of personalized firearms to further develop the technology, organize tests and demonstrations of reliability and effectiveness, and then based on the results of such tests, declare which models met the standards and would be eligible for federal procurement.  This is all within the standard authority of the executive branch.

Those models would then be made available as an option to federal officers who carry firearms.  They would not be required, but with over 130,000 federal officers carrying firearms, there will certainly be many who would prefer them.  Similarly, such models would be made available to the over 750,000 sworn state and local officers who carry firearms, in those jurisdictions that approve and for those officers who would prefer the safety of such arms.

Experience would then build confidence in the suitability of personalized firearms.  As that confidence grows, and as production costs come down with mass production, demand for such personalized firearms would also grow among private individuals.  Many gun owners – should they feel they have a continued need to keep guns in their homes – would prefer such arms.

In parallel, states keen on addressing gun violence would introduce liability insurance requirements.  Insurance in the US is organized at the state level.  Certain states could take the lead to establish the model and show what works, but the more states that participate the better.  The lower premium rates on the lower-risk personalized firearms would be a strong incentive to switch to such arms.

Ideally this should apply in all 50 states.  Realistically, one must recognize that a number of states will be reluctant or even opposed.  But as a track record is established, attitudes will hopefully change.  It may well take decades, as those attitudes are driven by fear and fears can be strongly held.  But the aim is a virtuous circle, where progress in reducing gun violence leads to the measures spreading more widely, which in turn leads to a further reduction in gun violence.

It certainly will not be perfect.  There will be those who seek to evade the law, and deaths due to guns will certainly not end overnight.  But with the US as such an extreme outlier, it would not take much to do better.


A.  Introduction

The tragedy of shootings in the US has not stopped.  In May, an 18-year old in Uvalde, Texas, slaughtered 19 school children – aged 9, 10, and 11 – plus two of their teachers.  Following such tragedies – as also after the killing of 14 teens and three adults at Stoneman Douglas High School in 2018, and the killing of 20 children aged 6 and 7 as well as six adults at Sandy Hook Elementary in 2012 – there are calls for something to be done.  But opponents of measures to place limits on the easy availability of weapons in the US have always succeeded in blocking any serious measure that would put a real dent in gun deaths in the US.

While it is commendable that Congress did pass a new gun safety law following Uvalde, it was modest at best.  While accurately described as the most important gun safety measure passed since 1994 (when a partial and temporary – now expired – ban on sales of new semi-automatic assault rifles was approved), the compromises required in order to secure enough Republican votes to allow passage given the Senate filibuster rules limits what it will do.  There will be enhancements to background checks to include juvenile records; individuals will be allowed to petition courts to limit gun ownership of previous intimate partners who have been guilty of domestic violence (closing what has been called the “boyfriend loophole”, as earlier law applied only to spouses); plus it authorizes increased federal funding to the states to implement mental health and crisis intervention programs and to enhance school safety investments.

While such measures are positive and should be applauded, it is telling that even such a modest bill can be accurately described as the most significant gun safety measure in close to three decades.  But sadly, there is no reason to believe that such measures, while positive, will have much of an effect on the overall number of those killed each year in the US from firearms.  Over 47,000 died in the US in 2021 due to gun violence.  And while mass shootings are particularly horrific, the number who die each year in mass shootings will normally be in the dozens.  Most deaths come from the “routine” use of guns to shoot someone, and rarely make national news.

Far more certainly could and should be done.  This blog post will describe one such reform – a major one – that actually could have a substantial impact.  It also could have a greater than zero chance of being implemented.  It would be a market-based approach based on the principle of individual responsibility (which might appeal to those conservatives who believe in individual responsibility as well as market-based measures), plus the focus would be on providing individuals a new choice – and not an obligation – on the kind of guns they may choose to own.

Nor would it require new legislation to be passed by Congress.  The initial impetus would be actions that a president has the authority to take.  This would then be complemented with measures that can be approved at the level of individual states.  States that are most willing to enact measures to reduce gun violence could take the lead.  While it would be more effective with a regional grouping of states, and more effective still at a national level, one could start in individual states and then see it spread as experience is gained and as a track record is established.

There would be two elements in the measures themselves.  The first would be a broad introduction of personalized weapon technology (often also called “smart guns”) – where the only one who can fire the gun is the owner.  The second would be a financial incentive to switch to such guns.  The latter would be achieved by the requirement that those owning a weapon must buy liability insurance for that weapon, where private insurance companies would set their rates for such coverage to reflect their assessment of the risk of that weapon causing harm.  One should expect the rates to be low on a personalized gun but relatively high for regular handguns.

Neither of these proposals is new.  But I have not seen the two put together before.  And together they should be expected to be far more effective than either individually, as they reinforce each other in an important way.

This blog post will present how such a program would work and could be implemented.  The post will first discuss how personalized firearms would not only stop many of the deaths (such as of children) that arise with current weapons, but would also very importantly change the dynamics of how criminals and criminal gangs arm themselves.  There would be no value to them of a personalized gun that had been stolen, or obtained through a straw purchaser, as they would not be able to fire it.  This will be followed by a discussion of how liability insurance on firearms would work.  And the section following that will then look at how one might get there from where we are now.

All major reforms to reduce gun violence in the US have failed.  While there is good reason to remain skeptical, the US is such an extreme outlier compared to other nations that it should not take all that much to get to a better place than where the country is now.  As the chart at the top of this post shows, homicides from the use of firearms in the US (adjusted for population) are close to an order of magnitude higher than in any other developed, democratic, country.  The closest is Canada, with a homicide rate due to firearms of 0.49 per 100,000 (based on the annual average from 2016 to 2019), versus 4.2 for the US in this period.  That is, the homicide rate was close to nine times higher in the US than in Canada.  It was more than twenty times higher than in the European Union, and well more than 100 times higher than in the United Kingdom.  [The figures were calculated from data collected by the Institute for Health Metrics and Evaluation (IHME), where the most recent comparable data across countries is for 2019 (and 2020 would have been a special case anyway due to Covid).  I took the annual average over 2016 to 2019 to avoid the possibility of an exceptional figure for some country in any given year.]

The incredibly high rate of gun homicides in the US could well be seen as depressing.  Why should the US stand out in this way?  But one can also see it as an indicator of what is possible.  Other countries from around the world show that deaths from guns at the rates seen in the US are far from inevitable.  It really should not take much to reduce US rates to well below where they are now.

B.  The Impact of Personalized Gun Technology

The aim is simple:  Make available guns that will fire reliably and with no special action by the owner, but not by anyone else.  That is, a positive identification would enable the gun to be fired by the owner, but not when someone else is handling the gun.  The technology exists, with alternative ways to enable this.  See, for example, here, here, and here, or this now somewhat dated 2013 report prepared by the US Department of Justice.

There are two broad approaches in the technology as it currently stands:  one relies on biometric information for the individual owner (with recognition of that individual’s fingerprints, palm prints, hand grip pattern, blood vein patterns in the hand, voice, and/or face), while the other uses token-based systems (where an RFID reader in the weapon is linked to an RFID tag or token embedded in an item carried by the owner, such as on a ring, watch, bracelet, wristband, and/or badge).  Token-based systems are the easiest to implement, but would not provide all of the advantages of a true personalized firearm.  While it would stop the accidental firing of a gun by a child that came upon such a weapon at home, for example, it would not stop such weapons from being sold to criminals in the black market (as the RFID token could be provided along with the weapon).  Still, it would be a step forward in reducing the risks of harm from firearms, and better than doing nothing at all.

While initially personalized guns will be basically hand-assembled and expensive, their ultimate cost should not be all that much more than for a regular gun once mass production starts.  ID systems no longer cost much.  Fingerprint ID sensors have been built into hundreds of millions of smartphones, as well as into devices such as laptops and keyboards.  And as for the cost, an Apple keyboard with Apple’s Touch ID button built-in costs only $50 more than a similar Apple keyboard without Touch ID (and this is at Apple’s premium prices).  While this technology, like any technology, can and should be further developed, technology is not the constraint on making such weapons now.

There are clear and important benefits.  Starting with the more obvious and then proceeding to those that would change the dynamics of how guns spread to criminals:

a)  Children:  It would save the lives of many children – from toddlers to teenagers.  All too often, young children come upon a loaded gun of their parents in their home, play with it, and then accidentally shoot themselves or a playmate.  As they get older, such guns may be used by a depressed teen or pre-teen to commit suicide.  And teenagers are all too often the victim of a firearm assault, usually by a handgun illegally obtained by another teenager.

In 2020, deaths due to firearms were the number one cause of death of those aged 1 to 19 – exceeding the number killed in traffic accidents.  And the mortality rate due to firearms in the US of children aged 1 to 19 – at 5.6 per 100,000 in 2020 – was close to 20 times the average in comparable countries of just 0.3:

This analysis comes from the Kaiser Family Foundation (KFF), using CDC and IHME data.  It is tragic that the US should stand out in this way.  While these figures include deaths when it was an adult handling the gun and pulling the trigger, such cases are only a small share of the total.

Rather, those deaths will largely be deaths due to other children holding the guns.  But by federal law, the minimum age to buy a handgun is 21.  Children cannot buy them.  While 18 and 19-year olds can buy shotguns and rifles in many (but not all) states – even though they cannot buy a beer in many of those same states – deaths due to homicides or suicides are largely from handguns.  According to FBI data, 94% of homicides by firearms (when the type of firearm was reported) were by handguns.  And almost all suicides are by handguns.  Thus almost all deaths of children due to firearms were likely by firearms obtained from someone else.  If children were not able to fire a gun they came across in their parent’s nightstand or obtained from someone else – due to personalized gun technology that would keep them from firing – most of those lives would have been saved.

b)  Police:  Two studies, of the partially overlapping periods of 1996-2010 and 2003-2013, each found that approximately 10% of the police officers who were killed while in the line of duty were killed by their own service weapon (or that of a colleague also at the scene).  This would typically involve a struggle with the suspect as the police are trying to subdue him (or her, but almost always a him), where the suspect was able to take control of the officer’s weapon and then use it on the officer.

This would not be possible if the police officers were using personalized weapons that only they could fire.

c)  Private individuals:  While there do not appear to be good statistics readily available on the number of private individuals who die from their own firearm when confronting a robber, burglar, or some other criminal (at least from what I have been able to find – only anecdotal stories such as this one), the share could well be higher than is the case when police officers confront criminals.  Police officers are trained to handle such situations; private individuals are not.

d)  Guns with Teachers in Schools:  While certainly not something that I would endorse, there are those (such as the NRA and former President Trump) who believe the way to make schools safe would be to provide teachers with loaded and easily accessible guns in their classrooms.  But one can easily imagine the tragedy that could follow when a disgruntled child or teenager could grab such a gun as easily as the teacher could.  And with millions of classrooms in the US, such tragedies would likely not be rare.

The situation would be different if it were a personalized weapon that could not be fired other than by the teacher.  While still not something I would recommend – teachers are not trained police officers – at least such guns would be of no use to a disgruntled student.

e)  Ending the trade in stolen guns:   The benefits listed above go to specific groups – certainly important groups (children, police officers, individuals confronting criminals, and teachers) – but still relatively narrow.  But with biometrically-based personalized guns the norm, there would also be a fundamental change in how guns spread in the nation, and in particular in how they get into the hands of criminals.

Specifically, since biometrically-linked personalized guns can only be fired by the owner, they will be of no use to anyone else.  Thus there will no longer be any reason to steal them, whether from an individual or from a gun shop.  This has become particularly important with the recent (June 2022) Supreme Court decision overturning a 109-year old New York State law that limited the carrying of concealed weapons outside of the home.  Guns carried outside of the home are easier to steal – with most of those thefts from unlocked parked cars.

We know what will likely now follow, as one can see what followed after a number of state legislatures chose to revoke such laws that had been on the books in their states.  A recent careful study (issued by the NBER, and summarized here and here) found that such changes led to a 35% increase in gun thefts, a 32% increase in armed robberies, and a 29% increase in overall firearm violent crimes (all controlling for other factors).

With personalized weapons, there would be no value in stealing a gun.  It would not work for the thief.

f)  Straw purchases:  With only biometrically-based personalized guns available, there would be no value in someone using a straw purchaser (such as a colleague or girlfriend) to purchase a gun when they were barred from doing so (for example due to a criminal record).  Such a gun would be of no use to them.  Such straw purchases would of course only totally cease when all guns for sale were such personalized guns.  Until then, gun manufacturers and gun shops would be serving two markets:  One for those who desire a gun only for their personal use, and one for guns that are of value to criminals who wish to illicitly pass along a gun they purchased to someone ineligible to buy one.

Such straw purchases serve not only criminals in the US.  The drug cartels in Mexico largely arm themselves by such straw purchases in the US, particularly in states with lax laws on gun purchases.  The Government of Mexico estimates that more than a half million guns are smuggled each year from the US into Mexico arming these gangs (with those guns largely obtained through straw purchases, they note).  They estimate that 70 to 90% of the guns recovered at crime scenes in Mexico had been smuggled from the US.

g)  Guns in gangs:  With biometrically-based personalized guns, there would be no point in passing around guns within a criminal gang to other members:  they would not be able to fire them.  But as long as guns that anyone can fire are available, criminal gangs will be able to obtain guns they can provide to other members for some criminal activity.  As in the case of straw purchasers, gun manufacturers and gun shops would know that they are serving two markets:  One where the guns are for personal use and one where they are of value to criminals.

h)  Black market:  Similarly, a black market in guns – for sale to individuals who would not be permitted to buy them from gun shops for some reason – could only exist for non-personalized weapons.  A personalized gun linked biometrically to an individual would be of no use.

Those who believe that weapons should be kept away from criminals should strongly support a move to personalized guns.  The NRA itself has noted:  “Most people sent to prison for gun crimes acquire guns from theft, the black market, or acquaintances.”  It went on to say: “Half of illegally trafficked firearms originate with straw purchasers who buy guns for criminals.”

Personalized guns would end this.

C.  Liability Insurance on Guns

The complementary reform to making personalized guns available would be to require that all gun owners carry liability insurance to compensate those harmed by the use of that gun – whether criminal or accidental.  The proposal was set out by Jason Abaluck and Ian Ayres, both professors at Yale, in a column written for the Washington Post and published in June 2022.  As they note, all car owners are required to carry liability insurance on their cars, to compensate those who may be harmed when that car is driven.  Those harmed by the use of a gun should be similarly compensated.

The basic elements of the approach are:

a)  The insurance would be market-based.  Private insurers would provide the policies, and would charge insurance premium rates that in their estimation reflect the risk that that gun might be used in a way that causes unjustifiable harm.  The harm might be accidental or criminal, but a victim has suffered either way.

b)  The amount of the compensation will depend on the extent of the harm, but would be paid at given amounts pre-set by a regulator (where in the US, insurance regulators are state-level entities).  This would be similar to how claims under workers’ compensation insurance are handled, where each state sets out a formula for the compensation based on the nature of the injury and other relevant factors.  There would then be no need to resort to the courts to determine damages in each case, thus avoiding the high costs and often long delays of the courts.

c)  The premium rates would then be set by the private insurer, in competition with other private insurers, at rates that best reflect that insurer’s judgment that their costs would be covered.  There would be a strong incentive for the insurer to research what factors are associated with the harms caused by the possession of guns.  One might speculate that these would include factors such as the age of the owner, gender, whether they had a criminal record or not, and more.  But one factor that some fear might be included would not be:  There could be no differentiation based on race (whether one thought – when other conditions are already taken into account – that this might be an independent factor or not), as that would violate anti-discrimination laws.

d)  The premium rates would also, and importantly, depend on the nature of the gun.  Hunting rifles are seldom used in crimes, and the insurance rates on such weapons would be low.  The rates on personalized guns – which only the owner can fire and hence are not ones that might be fired accidentally by a child or of any value to a criminal gang if stolen (as discussed above) – would be particularly low.

e)  While the actual premium rates would depend on what the insurers find by their due diligence, plus also depend on the standard rates of compensation that would be set in a particular state, to illustrate one might assume an average rate for such liability insurance on a gun might be $100 a year.  But on a hunting rifle it might be $50 a year on average, and on a hunting rifle of an older owner (say over the age of 30) it might be just $30 a year.  The liability insurance on handguns would cost more, as would insurance for those who are young.  This is all similar to what one finds with car insurance, where the rates depend on the age of the driver, gender, prior driving record, and the particular make and model of the car.

And it is important to note that insurance rates on personalized guns would be well below what they would otherwise be for the type of gun and for given factors such as the age of the owner – probably at least half or less.

f)  Liability payments in the case of suicides would need to be treated differently.  Suicide through the use of a readily available firearm is, sadly, common in the US.  Suicide by firearm accounted for over 60% of all firearm deaths over the last decade according to CDC data (in the decade to 2020 – the most recent year in the CDC data).  While it fell to “just” 54% in 2020, that was only because homicides due to firearms jumped by 35% in 2020.  (And it is noteworthy that despite all the stresses of Covid in 2020, the number of suicides by firearms in 2020 was basically the same as in 2019, and below what it was in 2018.)

If, in a suicide, the compensation payments were paid to the family of the deceased, there could be a perverse incentive motivating at least some of the suicides.  This is similar to the issue with life insurance, which is therefore not paid in cases of suicide.  To handle this, Abaluck and Ayres propose that payments from the insurer would still be made.  This would motivate the insurer to charge higher premia in cases where suicide risk may be high (which would in turn lead to fewer guns in the hands of those with such risk).  But the compensation paid would then go into a general fund to compensate victims of gun violence in cases where the guns used could not be traced.

Sadly, with suicides accounting for well over half of all the deaths due to firearms, this fund to compensate those where the firearm could not be traced would be “well funded”.  And this would address the criticism that some might have that the guns used in at least some of the criminal homicides might not be found.  It would still be possible to compensate those victims.

g)  All firearms would need to be registered, but this is already required (at least for certain types of firearms) in some states in the US.  While there are currently only six such states that require some such form of registration, those six states include California and New York and together account for one-quarter of the US population.  And it is worth noting that five of those six states (including California and New York) are in the top seven states in the US with the lowest death rates per capita from firearms in the US (CDC data for 2020).  The average death rate from firearms in those six states is less than half the rate in the rest of the nation.  While there are of course many factors involved in determining mortality from firearms, this does suggest – contrary to what the NRA would say – that such registration requirements do not make things more dangerous.

The registration process could also be used as the process by which a personalized firearm is “locked-in” to a specific owner.  That is, any newly purchased firearm would need to be brought in to some designated location (possibly some specific police station, as set by the local jurisdiction), that would have the special equipment needed to unlock the firearm and then lock it in to the fingerprints or other biometric measure of the specific owner.  Until it is unlocked in this way, the weapon could not be fired by anyone.  And after it is locked in to the new owner, it could not be fired by anyone else.

h)  Firearms sold in the US are already required by federal law to have a unique serial number engraved, stamped, cast, or otherwise embedded into them, to certain specifications (e.g. their depth) so as not to be easily removable.  Those serial numbers link a specific gun to a registered owner.  One would of course expect criminals would seek to evade this by trying to obliterate the serial number.  However, that is not easy.  Scratch marks are obvious, and even when a serial number is scratched out, there is special equipment that can often still recover the serial number, as the process of imprinting that number will lead to identifiable deformities in the underlying metal.

Police would in any case be legally able to seize any gun with a serial number that had been scratched out or otherwise hidden, or when not in the possession of the registered owner with a valid liability insurance cover – when the police find such a gun on someone who had been stopped for some reason.  Police are often blocked from seizing such weapons now and from making arrests connected to them.  Responsible owners would have no reason to worry, as there would be an ID on the gun and they would have proof of their liability insurance.  This is just as is typically required now when a car driver is stopped by the police for some reason (where the first request typically made by the police officer is to please provide proof of registration and insurance).

Criminals without this would indeed have reason to be concerned, just as a criminal driving a stolen car who is stopped by the police has reason to be concerned.  Others should not be.

i)  A more recent technology that is now available is for guns to place a unique, identifiable, mark on the cartridge casings that are ejected as a gun is fired.  Called microstamping, California passed a law in 2007 that all new models of semiautomatic pistols for sale in the state would need to incorporate microstamping once certain patents had expired so that the technologies would become publicly available without constraint.  Those patents expired in 2013, and hence the requirement came into effect from that date.  However, under pressure from the gun lobby, gun manufacturers then avoided the requirement by not introducing new models of such guns – as the law only applied to new models.  California then passed, with effect from July 1, 2022, a revised law easing the requirement and with other modifications, but it is too early at this point to see whether it will be skirted as well.

Microstamps on cartridges would prove to be of great help in tracing the guns used in a crime, and hence in solving those crimes, as spent cartridges will normally litter the scene.  Pro-gun groups have, however but not surprisingly, criticized the California law as unworkable, and brought lawsuits against it.  They argued that criminals would seek to evade the law by grinding down the mechanisms imprinting the microstamps.  However, the California Supreme Court ruled against them.  (Furthermore, while criminals do indeed seek to evade the law, that would seem to be tautological.  That is what criminals do.)

There is no doubt that the technology for microstamping, already good, could be made better.  That is true of any new technology.  But while there should be further such work, the technology as it exists would already be of tremendous benefit in solving many of the unfortunately numerous cases where guns are fired in a crime in the US.  And while such microstamping would lead to more criminal cases involving guns being solved, it is not absolutely necessary for a liability insurance system to work.  Rather, it would make such a system work better.

It does, however, pose a separate issue.  One would want to encourage all responsible gun owners to arm themselves – in those cases where they feel they have a continued need to arm themselves – with weapons that incorporate such microstamping technology.  Many of those weapons are subsequently stolen or otherwise end up arming a criminal, and if they are not personalized weapons, may be used in a crime.  Then, precisely because it would be possible to identify the weapon that had been fired to harm somebody even if the shooter got away with his gun, the insurer who had provided liability insurance on that weapon would need to pay the resulting claim.  Because of this, insurers would have an incentive to charge higher insurance premium rates on weapons with the microstamping technology.  This would not apply to personalized guns – as such guns could not be fired by anyone other than the owner – but would apply to traditional weapons.

To address this, one could have a requirement from the state’s insurance regulator (or by legislation) that the insurance rates on weapons with the microstamping technology would be charged at some lower rate – perhaps half – than the rate for a similar weapon but without that technology.  To make up for the other half, transfers could be made from the fund discussed above to compensate victims where it was impossible to trace the gun that was responsible for the harm.  That fund would be funded by the compensation that would have otherwise been paid in cases of suicides, and with suicides in recent years generally accounting for over 60% of deaths due to firearms, that fund would have ample balances.

j)  Requiring liability insurance coverage would not only be a strong incentive to choose a personalized gun if one is going to buy a gun, but would also be a strong incentive to turn in at least some of the extra guns that a person may have.  According to a survey by the Pew Research Center, 66% of those who own a gun say they own more than one gun, and 29% say they own five or more guns.  Gun owners, when faced with paying liability insurance premiums each year for each of their guns, would now have an incentive to turn in their excess guns (possibly all of their guns) in one of the increasingly common police buy-back programs, rather than simply let them accumulate in their home.

As they grow older, gun owners will also often realize that guns around the home – especially several guns around the home – do not make their families safer.  But it is easy now to do nothing and allow such guns simply to lie around in a cabinet.  Having to pay an insurance premium each year would be a reminder, as well as an incentive, to do something about it.  And taking such guns out of circulation would end the risk that they may end up in someone else’s hands and be misused.

k)  Insurers providing liability insurance on firearms would require the loss or theft of a firearm they insured to be reported immediately to the police.  There is no nationwide practice on this now, other than by federally licensed gun dealers (where there is a federal law requiring this).  While 13 states plus Washington, DC, do require such reporting by gun owners, and two more require such reporting in certain circumstances, the 35 other states have no such requirements.  While this reporting would not be as important for personalized guns (as discussed above), stolen guns that can be fired by anyone feed criminal activity.

Mandatory liability insurance would be an additional incentive to be careful where one stores a gun.  Guns kept in unlocked cars are a major source of such thefts.  Someone with a history of guns being lost due to such negligence would pay higher liability insurance rates.  This would lead gun owners to be more careful.

l)  Insurance requirements are a state-level matter in the US.  Thus states can decide individually whether to require liability insurance on firearms kept by residents of their states, just as each state decides what to require on liability insurance on cars.  Note there is no Second Amendment issue here – even though gun advocates will undoubtedly claim that there is.  While the modern interpretation of the Second Amendment by the current Supreme Court is that there is an individual right to keep a firearm at home (and with the recent decision overturning the 109-year old New York State law, also a right to carry concealed weapons outside of the home), there is no right to getting such firearms for free.  You must pay for them.  Similarly, a state can require you to pay for liability insurance that will compensate those harmed by an illegal or accidental discharge of that firearm.

It would of course be far better for all states to require such liability insurance rather than just some.  But an individual state could adopt it, and ideally work with other nearby states so that such insurance would be required on a regional basis.  Those living in reluctant states that do not at first have this would then see how well the process has worked in the states that had adopted it.  Assuming it proves effective, voters in those states – “armed” with such evidence – can then work to require their own legislatures to adopt similar measures.

This now gets into the process by which these dual reform measures (personalized firearms and liability insurance on firearms) could be implemented – the topic of the next section below.  We have seen that the reforms are workable.  There is nothing impossible here.  The issue, rather, is one of willingness.

D.  A Step-by-Step Program

How to get there from where we are now?  It will not be easy, and one should recognize that it may take decades due to the emotions involved.  Attitudes will need to change, and as these attitudes are driven by fear, they can be strongly held.  Current gun owners will need to see that personalized weapons are reliable and effective, as well as safer for them and their loved ones.

Proceeding step-by-step would allow confidence to build over time, fears to diminish, and should ultimately lead to a far better place than where we are now.  A program can start with measures that the Biden Administration, acting within the authority a president has, can act on.  States willing to take measures that would curb gun violence can also act.  Experience can then create comfort, and demonstrated progress can lead to confidence that this path forward leads to lower gun violence.  The aim is a virtuous circle, where progress in reducing gun violence engenders greater confidence and hence willingness to implement such measures more broadly, which in turn leads to lower gun violence.


a)  To start, the Biden administration should work with both current manufacturers of personalized firearms, as well as with potential new manufacturers, to further develop the technologies involved.  While technologies exist, they can always be refined and improved – making them more reliable, easier to use, and lower cost.  This can be done with the authority federal agencies (such as the FBI, the ATF, and others) have to develop equipment they might use.  A focus would be on reliability – that the personalized gun will always fire when the finger of the owner is on the trigger and not when that of someone else is.  One would also want a technology that could not be disabled or disconnected – where any attempt to do so would lead to a weapon that could not be fired at all.  And there are critics who assert that a personalized gun could be “hacked” remotely.  I do not see how this would be possible, at least for a fingerprint ID or similar biometric system, as it would be totally self-contained to the gun and not connected to the internet or some other external system.  But to address such possible concerns, one would also want to show the technology developed cannot be hacked in some way.  Critics could be invited to test the systems themselves.

b)  As such technologies are developed, the Biden administration can then organize regular evaluations of the technologies to test (and demonstrate) how well they work, and to determine also where further development work might be focused.  Based on such testing, they can then announce which specific personalized firearms would be eligible for federal procurement.

c)  With such personalized weapons developed, tested, validated, and declared eligible, they would then use standard federal procurement procedures to make available such firearms to federal law enforcement officials who wish to arm themselves with such weapons.  This would be voluntary, but with over 130,000 federal non-military officers authorized to carry firearms, there would certainly be many who would prefer the safety of such personalized firearms.  Armed federal officials include not just those in well-known agencies such as the FBI and the Secret Service, but also those in, for example, the border patrol and federal prison guards.  Many would prefer a weapon that could not be used against them in case they get into a struggle with someone they are trying to arrest or control.

d)  There are also about 750,000 sworn state and local law enforcement officers in the US, spread over about 18,000 government agencies.  Sworn law enforcement officials carry firearms.  There are an additional more than 300,000 non-sworn officers in such agencies, some of whom carry firearms for their job.  Many of these officers would choose to use personalized firearms if given that option.  They should be given that option.

e)  While the number of law enforcement officers who choose personalized firearms over traditional ones might at first be relatively small, that number can be expected to grow over time as the personalized firearms prove their reliability and safety.  And one should expect jumps in those choosing such weapons following publicized cases of law enforcement officers being shot with their own weapons.

f)  Based on the experience gained by law enforcement officials who have voluntarily chosen to carry personalized firearms, as well as with the continued development and refinement of the technologies involved, one could move over time to wider use of such weapons in these agencies.  As confidence is gained in their reliability and advantages, personalized weapons could become the standard weapons issued in at least certain federal agencies.  Federal financial assistance to state and local law enforcement agencies could similarly be geared to encouraging the use of such weapons.  Such federal financial assistance is significant, and could, for example, fund a higher share of the costs when the procurement is of personalized weapons than when it is for traditional firearms without such protections.  Again, the pace of the shift would depend on a demonstration of reliability and as confidence is gained.

g)  With the development of what would be a substantial market for such personalized firearms for law enforcement officials at the federal, state, and local levels, and the confidence that such use would engender, the much larger private market could develop.  With demonstrated effectiveness, as well as falling costs as mass production becomes possible, one would see interest in such personalized weapons among private individuals who believe they need a firearm for their personal protection.

h)  In parallel, states that are most interested in reducing gun violence could take the lead in adopting measures to encourage the use of personalized weapons.  This would start with mandatory registration and reporting requirements, which some states already require (with this associated with lower per capita deaths from guns than in other states – although this is likely due to many factors).

i)  Such states would also start to require that individuals take personal responsibility for any unjustified harm that might be caused by their weapons, by carrying liability insurance on whatever guns they own.  The states would set standard compensation amounts for those injured by the use of a gun that causes harm – whether accidental or criminal.  Private insurance companies would set the premium rates for such insurance, in competition with other private insurers.  As discussed in Section C above, such insurance rates can be expected to be far lower for personalized firearms than for similar firearms that can be fired by anyone.  This would provide a further incentive for individuals to choose such firearms over traditional ones.

j)  And also as discussed above, this would be complemented with generous buy-back programs designed to get as many traditional guns out of circulation as possible.  These programs already exist in many jurisdictions.  They would become particularly valuable as personalized firearms replace traditional firearms, and as individuals take on the costs (through the liability insurance they would be required to obtain) that they now impose on others from the harm caused by such weapons.

k)  There may also be transitional issues to be addressed.  If only a few states require liability insurance, a high share of the victims of gun violence in those states may well be from guns obtained by criminals in other states.  When the guns used could not be traced, those harmed by those guns (or in the case of homicides, their estates) would be compensated from the general fund discussed earlier (funded by payments that would otherwise have been paid in the cases of suicides).  When the guns can be traced (as they have serial numbers on them) and came from a state that does not require liability insurance, the question would arise whether there might be some other process – possibly via private lawsuits – to hold those responsible accountable for the harm caused by those weapons.

But one should not over-complicate this.  First, such transitional issues might not, in practice, be all that important quantitatively.  The magnitudes will depend on several factors.  And it will matter less as more states sign on to such a program.  The sooner they do, the better.  But if, to start, only a few states participate with then a high share of the harm resulting from guns obtained from other states, it is possible that the compensation paid to the victims might have to be limited.  But even limited compensation is better than no compensation at all, which is what we have now.

E.  Conclusion

Serious reform measures that would reduce gun violence in the US have repeatedly failed in recent decades.  As a result, even the most ambitious programs now being proposed are so modest that few believe that, even if approved, they would have a significant impact on the overall numbers.  They are still worthwhile – as any death averted is still of value.  But no one believes that a serious reform program would ever be passed by the US Senate, where just 40 senators can block any action.

What has been suggested here is an alternative approach.  One would proceed step-by-step, starting with actions the president can take on his own authority.  Individual states can then also proceed with measures they have the authority to pass and implement.  At least certain states, including several of the larger ones such as California, New York, and Illinois, would likely be willing.  Over time, as experience is gained and the benefits become clear, other states should be expected to join.  And at some point, even the most reluctant states would have to recognize that their lack of action is only serving to arm criminals.

The Biden administration should presumably have no issue with proceeding as presented above.  The Biden campaign platform on gun safety issues said specifically:

Put America on the path to ensuring that 100% of firearms sold in America are smart guns.  Today, we have the technology to allow only authorized users to fire a gun. For example, existing smart gun technology requires a fingerprint match before use. Biden believes we should work to eventually require that 100% of firearms sold in the U.S. are smart guns.

Given this commitment, the Biden administration should have no issue with ordering work to be done to further develop, test, and validate personalized firearms, and then to make such weapons available to federal law enforcement officials who carry a gun as part of their job.  I am not aware, however, of any actions along these lines that are underway.  A “Fact Sheet” released by the White House on July 11, 2022, listed 21 executive actions that President Biden has taken to reduce gun violence, but smart guns were not mentioned.

One can be certain that the NRA would still oppose this.  The knee-jerk reaction of the NRA in recent decades has been to oppose any and all gun reform proposals.  But at least formally, the NRA is not opposed to the development of smart guns nor to making them available for gun owners to acquire.  They say they are just opposed to making them mandatory.  Their position (as recorded on their website) is:

The NRA doesn’t oppose the development of “smart” guns, nor the ability of Americans to voluntarily acquire them.  However, NRA opposes any law prohibiting Americans from acquiring or possessing firearms that don’t possess “smart” gun technology.

The proposal as presented above is fully in line with this.  Smart guns would be developed and made available, but not made mandatory.   The NRA should in principle not be opposed.

But the NRA’s history on the issue is not encouraging.  In 2000, Smith & Wesson (the oldest and largest gun maker in the US) announced that, working with the Clinton White House, it would introduce a smart gun to the US market.  The NRA was outraged, and with its gun owner allies organized a boycott of Smith & Wesson firearms that drove the company close to bankruptcy.  The firm was sold to a new owner (for just $15 million) with the new owner reversing the plans.  The boycott ended, and Smith & Wesson once again became one of the largest manufacturers of guns in the US.  Not surprisingly, the firms planning now to bring smart guns to the US market are all start-ups.  They are not among the major gun manufacturers, who would be vulnerable once again to an NRA-led boycott.

What is key is that the NRA recognize that we are all on the same side here.  We all agree that we do not want “bad guys with guns” to commit crimes.  And should an individual wish to purchase a firearm, they will remain eligible to do so under this proposal.  An increasing share should, over time, see the advantages of personalized firearms over those without such safety mechanisms.  But it will be many years before the nation gets to the point where 100% of the new firearms being sold will incorporate such safety mechanisms.

It will not be perfect, certainly, but there is a need to start.  Any lives saved are worthwhile.  And the Biden administration can take those critical first steps.

Measures of GDP; How Recessions Are Determined and Dated; the Economy in the First Half of 2022; and the Prospects for 2023

A.  Introduction

The Bureau of Economic Analysis (BEA) of the US Department of Commerce released on August 25 its second estimate of the GDP accounts for the second quarter of 2022.  The figures indicate that GDP fell by 0.6% in the quarter, a bit less than the fall of 0.9% in its initial estimate released in late July (what it calls its “advance estimate”).  But it was still a fall, and following the reduction in GDP in the first quarter of 2022 (by 1.6% in the most recent estimate), there have now been two consecutive quarters where estimated GDP has gone down.

Many mistakenly believe that an economic recession is defined as two consecutive quarters of falling real GDP.  This is not correct – there is no such definition for a recession.  But it is easy to see that such confusion can arise, as a commonly used “rule of thumb” is that if real GDP fell for two consecutive quarters, then this is a sign that the economy is in a recession.

The reality is more complex.  Much more enters into a designation that the US economy was in a recession in some period.  Indeed, while the quarterly GDP figures are certainly important, they actually play a secondary role as the designation of a recession is based more on a number of indicators that are available on a monthly basis (such as the monthly employment figures, wholesale and retail sales, and more).  Indeed, the dates assigned to a recession (when it began and when it ended) are of specific months, not calendar quarters.

Usually this does not matter much.  Such economic indicators normally move together.  But not always, and they certainly have not in 2022 thus far.  While real GDP as currently estimated fell in the first half of this year, the employment market has been extremely strong.  Employment has grown by an average of over 440,000 per month in the first half of 2022, and the unemployment rate fell from an already low 4.0% in January to just 3.6% in June and an even lower 3.5% in July.  This is the lowest the unemployment rate has been since 1969 – matching the 3.5% rate hit in early 2020 just before the pandemic crisis.  While a formal determination has not been made on whether the economy is in a recession or not – and as discussed below will not be made until more of the data are in and the trends are clear – it is highly doubtful that the first half of 2022 will be so designated.

This blog post will cover how that designation process works.  But it is of interest first to look at the current estimates of what has happened to real GDP in the first half of 2022.  The period illustrates well the pitfalls of exclusively focussing on whether real GDP fell for two consecutive quarters as an indicator of whether the economy is in a recession.

There is indeed a question of whether GDP in fact fell in the first two quarters of 2022 – even setting aside the issue that there will be further revisions in the current estimates.  Specifically, the BEA issues figures for GDP based on two different ways of estimating it:  One is based on expenditures (for consumption, investment, etc.) which it labels the expenditure-based GDP (or just GDP for short), and another is based on incomes earned (which it labels Gross Domestic Income, or GDI for short).  They should in principle be identical, as whatever is spent is someone’s income.  But the two estimates will differ in practice, as they are based on different approaches and different sources of data.

As seen in the chart at the top of this post, these two measures of GDP, while generally moving together over time, have diverged significantly from each other since late 2020.  And in the first half of 2022, GDI continued to grow while GDP fell.  The reasons for this divergence are not clear, but I am sure economists at the BEA are ardently trying to figure this out now.

At this point we do not know what the answer is.  It might well simply be a consequence of the estimates still being recent, and might go away as further data become available to yield better estimates.  But that difference between the two estimates illustrates well why one should not simplistically assert that two quarters of real GDP decline signals a recession underway.

This post will thus first look at the recent data, focusing on what the GDP and GDI concepts mean, why they should be identical (and indeed, for this reason serve as a useful check on each other in the estimates), and what might have caused the recent divergence.  The post will then look at the process followed in the US for designating periods of economic recession and expansion, where for historical reasons the process is overseen not by the government, but rather by a nonpartisan organization called the National Bureau of Economic Research (NBER).  It will conclude with a brief discussion of the prospects for 2023.  While it is doubtful that the economy in the first half of 2022 will ever be designated as being in a recession, the prospects of a recession in 2023, or even later in 2022, are substantial.

B.  Gross Domestic Product and Gross Domestic Income

Gross Domestic Product (GDP) is a measure of production – how much the economy is producing.  But while it is a measure of production, the primary way estimates are made of how much was produced, as well as the way most people think of GDP, is not by how much is produced but by how much is used.  That is, everyone who has taken an Econ 101 macro course will know that GDP will equal the sum of Private Consumption, Private Investment, Government Consumption and Investment Spending (often combined as simply Government Spending – but excluding spending on transfers to households such as for Social Security), and Net Foreign Trade (Exports less Imports).

Why should that sum of expenditures equal production?  The trick (as discussed in this earlier post on this blog) is that investment includes investment in any net buildup of inventories.  That is, changes in net inventories in a period will balance out any difference between what was produced and what was sold.

This is then a convenient way to estimate GDP.  But one should keep in mind that GDP is a measure of production, and that there are other ways to measure that which should yield the same result.  One is to approach it via incomes, as whatever is produced and sold is then someone’s income (when one includes the value of any net inventory accumulation).  Those incomes accrue as someone’s wages (including all forms of labor compensation) or as profits (net operating surplus more formally).  The BEA can assemble available data on wages and profits in the economy, and the sum should in principle be the same as GDP (with adjustments for indirect taxes such as sales taxes and including whatever was set aside in depreciation allowances).  (For those interested in the detailed breakdown, see Table 1.10 in the BEA NIPA Interactive Tables.)  For clarity, the BEA labels this income-based estimate of what should sum also to GDP as Gross Domestic Income, or GDI.

A third approach to estimating GDP is to estimate directly what production was in each sector of the economy.  The BEA does this as well, but one needs to take into account that the net contribution to production in the economy as a whole is not the gross output of any given sector, but that gross output less the value of whatever inputs it purchased from other sectors of the economy.  This is so that one does not double-count what is being produced.  That is, in each sector one estimates what economists call “value-added” – the value of what was produced less the value of the material inputs purchased to make that product.  The sum of this value-added across all sectors should once again be GDP.  The BEA refers to these estimates of value-added by sector as “GDP by Industry”.

The three measures should in principle yield the same figures for overall GDP.  But while in practice generally close, they don’t exactly match as they are all estimates based on data, and the data come from different sources.  Furthermore, that data is subject to revision as more complete information becomes available, so even though initial estimates may differ by some amount, the degree of those differences generally falls over time as better estimates become possible.

Why then does the public discussion generally focus on the expenditure-based estimate of GDP?  One simple reason is that it is always the first one that is published.  The BEA issues this initial estimate of GDP (its “advance estimate”) just one month after the end of the calendar quarter.  This estimate is eagerly awaited both by policymakers and the general public, and receives a good deal of attention in the news media.

The BEA only releases its first estimate of the income-based estimate of GDP (i.e. GDI) a month later, along with its second estimate of the expenditure-based approach to estimating GDP.  Since it comes later, and possibly also because it is less well known, less attention is given by the public (and consequently in the news media) to this income-based estimate of GDP.  But the quarter-to-quarter changes in GDI can differ significantly from the quarter-to-quarter changes in the expenditure-based estimate of GDP.  For example, in the estimates released on August 25, the revised (“second estimate”) for expenditure-based GDP was of a fall of 0.6% in real terms (at an annual rate and seasonally adjusted).  However, the initial estimate of the income-based estimate of GDP (i.e. of GDI) was that GDP grew by 1.4%.  This will be discussed further below.

The initial estimates using the third approach to estimating GDP (i.e. value-added by sector) are then only made available a month after that, i.e. along with the third estimate of the expenditure-based estimate of GDP and the second estimate of the income-based estimate of GDP (i.e. GDI).  These estimates receive even less attention.  The BEA has also been publishing them along with the monthly GDP reports only recently – starting in September 2020 for the second quarter of 2020 GDP figures.  They released them separately before with some further lag, and the underlying data series themselves are only available (in a consistent series based on the current methodology used) from 2005 on a quarterly basis and from 1997 on an annual basis.

Furthermore, while this third approach to estimating GDP could yield an additional check on the GDP estimates, in practice the BEA does not do this.  I am not sure precisely why, but in its methodology for estimating these GDP by Industry figures, it scales the estimates so that the sum matches the expenditure-based estimate of GDP for the period.  The BEA may feel that the underlying data for the GDP by Industry estimates are not sufficiently good to provide an independent estimate of GDP, or it might be concerned that a third but different estimate for GDP might cause confusion in the public.

It is thus not surprising that most attention is paid to the expenditure-based estimates of GDP.  They are available first, and thus they provide the figures that first indicate whether GDP is rising or falling.  But there is also a more fundamental reason why they deserve such attention.  As we have known since Keynes, the primary driver of GDP in the near term is what is happening to the various components of demand for GDP, i.e. the expenditure-based components of GDP.  Production (within the bounds of productive capacity) will respond to those demands, and in particular production will fall when the sum of those demands (what economists call “aggregate demand”) falls.  This might be in response to some financial crisis (with chaos in the financial markets leading to less investment), or to the Fed raising interest rates with the deliberate intention of reducing demand (with the higher interest rates leading to less investment), or due to cuts in government spending (possibly due to politics, such as when the Republican-controlled Congress elected in 2010 forced through government expenditure cuts in the subsequent years, thus slowing the recovery from the 2008/09 financial and economic crash while blaming this on Obama).  Similarly, spurs to growth will be found in what is happening to the various expenditure components of GDP.

The interest in this estimate of expenditure-based GDP is thus well-founded.  But one needs to keep in mind that the figures are still estimates, and are imperfect as the data are imperfect.  An independent check on this, such as from the independent estimate of GDP based on estimated incomes (i.e. GDI), is thus of interest.  Henceforward, for simplicity I will generally refer to the expenditure-based estimate of GDP as simply “GDP”, and the income-based estimate as simply “GDI” (the same terms the BEA uses).

The two estimates (GDP and GDI) generally move quite closely together.  This can be seen in the chart at the top of this post.  Note that while the figures here are shown in real terms, the price deflator used for both GDP and GDI is the same.  The reason is that while price indices can be calculated for the goods and services that make up the expenditure-based estimate of GDP, one cannot define such price indices for the wages and profits that make up the income-based GDI. Thus to deflate the GDI estimate to real terms, the BEA uses the same price deflator as it has estimated for GDP.  This is convenient for the interpretation of the figures as well, as any deviation of one from the other cannot then be attributed in some way to two different price deflators being used.  There is only one.

[Technical Note:  The figures are of GDP and GDI each quarter, but they are shown at annual rates from seasonally adjusted figures.  The price indices used are what are called “chain-weighted dollars”, with 2012 as the base year.  One may recall from an Econ 101 class that a Laspeyres price index calculates the index based on the weights of the underlying items in overall expenditures in the base year, and a Paasche price index calculates the index based on the weights of the underlying items in overall expenditures in the final year.  A chain-weighted index calculates the index based on weights that change period by period based on expenditures on the items in each of the periods.]

The estimates of GDI have generally been above the estimate of GDP in recent years – and especially so since late 2020.  That has not always been the case.  One can see in the chart at the top of this post that estimated GDI was below estimated GDP between mid-2007 and the start of 2011.  But broadly they move together, as one should expect and as can be seen in a chart of the data going back to 1947 (when quarterly estimates of GDP and GDI began):

There is, of course, a scale effect over such a long period, as real GDP has grown by a factor of ten between 1947 and 2022.  The difference between GDP and GDI will not then be so apparent in the earlier years, and it is more meaningful to look at the difference between the two estimates as a share of GDP in that year:

The BEA assigns a label to the difference between GDP and GDI:  they call it simply the “Statistical Discrepancy”.  That difference as a share of GDP was quite small and generally within a range of +/- 1% of GDP between 1947 and the late 1970s, and more often positive than negative (i.e. estimated GDP above estimated GDI).  It then moved between greater extremes, but remained generally positive, from the early 1980s to around 1997.  The volatility then continued, but since 1997 the Statistical Discrepancy was more often negative than positive (estimated GDP less than estimated GDI).

Since the fourth quarter of 2020 it has, however, turned more sharply negative than it has ever been before.  Why?  No one really knows, although there is some speculation (and I am sure work underway at the BEA to try to figure this out).  A higher GDI than GDP implies that estimated incomes are higher than what the expenditure-based estimates would imply.  It is possible that some of these incomes are becoming more difficult to estimate.  For example, there are conceptual issues in how properly to account for compensation being paid by transfers of assets – such as happens with stock options – and the BEA data sources may not be good at estimating these.  Individuals may treat these as part of their compensation (as they should), but in the company accounts they may be treated as a transfer of assets (the stock options) that may not then be properly reflected in recorded profits (at least from the viewpoint of the National Income Accounts).

It is also possible that the sharp increase in the Statistical Discrepancy in the last couple of years may in part go away as more complete data becomes available and new and better estimates for GDP and GDI are worked out.  But at this point we just do not know.

Due to these differences in the estimates, many of the more careful economists working with the GDP figures use not solely the GDP estimate nor solely the GDI estimate, but rather the simple average of the two.  By weighting them equally in this simple average, the implication is that the uncertainty on each is similar.  The BEA itself provides this simple average in its monthly releases of the GDP estimates (although with the item blank in the first release of each quarter when only the expenditure-based GDP estimate is available).  But these figures on the average of GDP and GDI do not receive much attention from many.

Focusing in on the last few years:

The chart is as before, but now shows also the simple average of the GDP and GDI estimates.  The path of GDP as estimated by the GDI figures has been substantially above the path as estimated by the expenditure-based GDP figures since the fourth quarter of 2020.  And in the first half of 2022, GDI has continued to grow (although at a slower pace than before) while GDP as measured by expenditures fell.  Neither of the changes are large.  And the simple average of the two comes out as almost flat, but positive (with growth of 0.1% in the first quarter of 2022 and 0.4% in the second quarter – in the estimates as currently published).

Thus by this measure of GDP, the economy has continued to grow in these most recent estimates in the first half of 2022, although at only a slow rate.  This could well change with the revisions to come as more complete data become available, but for now they show positive growth in each of the quarters.

C.  Designating and Dating Recessions

The commonly accepted designation of whether the US economy is in a recession or not is not made by a government agency, nor is it based on some set of specific criteria (such as that GDP fell for two consecutive calendar quarters).  Rather, for historical reasons the designation is made through a private, nonprofit and nonpartisan, organization that supports economic research in the US called the National Bureau of Economic Research (NBER).

The NBER was founded in 1920, on the initiative largely of two individuals – one an executive at AT&T and the other a socialist labor organizer who had a Ph.D. in Economics from Columbia.  While very different in their views on what to do about unemployment, both recognized that the data available at the time were insufficient for an adequate understanding of the conditions.  They founded the NBER with the intention for it to support teams that could produce such data – more than what could be done by individual academics.  They deliberately kept it nonpartisan, where the NBER itself would not produce specific policy recommendations, and were able to obtain funding from a range of sources, including from some of the larger corporations of the time, from certain foundations, and from other private donations.

The NBER’s first director of research was Wesley Clair Mitchell, then a professor at Columbia and an expert on business cycle research.  He assembled a team that produced what was then the best data of the time on business-cycle fluctuations in the US.  This research was published and proved influential.  As part of it, as well as in continued such work later sponsored by the NBER, the researchers would determine, to the best the data they could assemble would allow, the periods when the US economy was expanding and when it was contracting.  Periods of contraction were labeled recessions.

The US Department of Commerce started to produce more systematic data on the state of the economy in the 1930s, due in part to the Great Depression then underway.  They worked out the basic GDP concepts we now use and how to measure them in practice given the data they could assemble, with this early work done often with the help of researchers from the NBER.  A particularly prominent such then-young researcher was Simon Kuznets, a student of Wesley Clair Mitchell who then moved to the NBER, and who is often credited with developing the original concepts for GDP (and who subsequently was granted a Nobel Prize in Economics for this work).

The Department of Commerce (now through its Bureau of Economic Analysis) has since produced the official GDP accounts for the US.  In 1961, a decision was made that rather than have this government agency make a determination on whether the economy was in a “recession” (defined in some way) or not, they would instead simply reference the determinations made at the NBER.

These determinations of the NBER were originally made as a by-product of the research it sponsored on business cycles in the US.  In 1978, the NBER decided to formalize the process and make it independent of specific research projects by appointing a committee of academic economists to make such designations.  The committee members represented a range of views but all members had a focus on macro and business cycle issues.  Formally, it was named the NBER Business Cycle Dating Committee.  There are currently eight members of this Committee, and there has been only limited turnover over time.  There have been only seven other individuals who have served on the Committee in the 44 years since its origin, and the chair (Robert Hall), as well one of the current Committee members (Robert Gordon), have served on it since its start.  Robert Hall is a well-respected economist, a professor at Stanford since 1978, and is politically and economically conservative.  He was a supporter of the Reagan tax cuts and has advocated for a flat tax to replace progressive income taxes in the US.

This NBER committee was set up by Martin Feldstein (a professor at Harvard) soon after he became the president of the NBER.  Feldstein was also a well-respected economist as well as open-minded.  He was the Chair of the Council of Economic Advisers in the Reagan White House between 1982 and 1984.  During that time he brought to the Council two bright and capable young economists with recent Ph.Ds. – one to look at domestic policy issues (Larry Summers) and one to focus on foreign trade issues (Paul Krugman).

The NBER Business Cycle Dating Committee meets when members believe they have sufficient data and other information to determine whether the economy had reached a business cycle peak (following which it would be contracting, with this then a recession), or a trough (after which the economy would be expanding, and the recession would be over).  Such determinations have been made by the Committee anywhere between 4 and 21 months after the dates of those business cycle peaks or troughs (as later determined).  They have no deadline for this, but meet when they believe they may have sufficient data to draw a conclusion.  Indeed, sometimes they have met and then deferred a decision, as they felt that upon review they did not yet have sufficient information to make a decision at that point in time (see this news release for one example).

Keep in mind that an economy in recession is one where economic activity is contracting.  It is not defined as a period where economic activity might be considered “low” in some sense, such as below some previous peak.  Thus unemployment will in general still be relatively high at the point where the economy has started to expand again and has thus emerged from the recession.  This may be confusing to some, as economic conditions “feel” (and in fact are) very similar to how they were the month before a trough was reached.  Indeed, it is common that the unemployment rate will still be growing for a period after that trough even though the economic recession (as defined here) is over.  For example, the NBER Committee determined that the 2007/2009 contraction (and thus recession) ended in June 2009.  At that point, the unemployment rate had hit 9.5% – higher than at any point since Reagan (when unemployment peaked at 10.8%).  But the unemployment rate continued to rise after June 2009, peaking at 10.0% in October 2009.

How then is a “recession” defined?  The NBER Committee defines it as:

“a significant decline in economic activity that is spread across the economy and that lasts more than a few months. The committee’s view is that while each of the three criteria—depth, diffusion, and duration—needs to be met individually to some degree, extreme conditions revealed by one criterion may partially offset weaker indications from another.”

Note that it must be what the Committee determines to be a “significant” decline, spread across much of the economy and not simply concentrated in a few sectors, as well as a decline that lasts for a substantial period (normally more than just a few months).  But no specific minimum values are specified for any of these factors.

The Committee also dates the recession (i.e. the dates of the peak and the trough in economic activity) to a specific month.  For this reason alone, the GDP data will not suffice.  It is only available quarterly.  Rather, the Committee has explained that it pays particular attention to the following data series (from the BEA, the Bureau of Labor Statistics, and other sources), which are made available and published monthly:

Real personal income less transfers;

Real personal consumption expenditures;

Employment (both nonfarm payrolls from the Survey of Establishments and employment as reported in the Current Population Survey of households);

Real manufacturing and wholesale/retail trade sales;

Index of industrial production.

But while the Committee has explicitly noted it pays attention in particular to these data series, they can and will look at whatever they feel may be relevant to their decision.

Once they determine the month in which the economy reached a peak or a trough, they will also report on which calendar quarter they believe the economy reached its peak or trough.  This is normally, but not always, the calendar quarter of the respective peak or trough of the months marking a recession, but not always.  Sometimes it might be the quarter before, or the quarter after.  For example, in the short but extremely sharp downturn in the spring of 2020 due to the lockdowns required to deal with Covid, the date marking the start of the recession (when the economy had reached its peak) was February 2020 and the trough was set as April 2020.  But the peak quarter was determined to be the fourth quarter of 2019, not the first quarter of 2020.

It also should be noted that for these determinations of the quarters where the economy had reached its peak or trough, the Committee does not focus on the expenditure-based estimate of GDP, but rather on the simple average of this GDP and GDI.  And as noted above, by this measure GDP rose in the first half of 2022 (according to the current estimates).

Could the Committee get this dating wrong?  Certainly – they are only human, and judgment is required in making these decisions.  Others can and sometimes do disagree, as one would expect in any science.  But the Committee has been careful, makes its decisions only when they believe sufficient time has passed to allow them to make a decision, and the members of the Committee represent a range of perspectives.  And while they do not say so explicitly on the NBER website where they explain their work, I strongly suspect that the Committee operates by consensus, and that if there is not a consensus when some such meeting has been called, they defer their decision until more complete data allows a consensus to be reached.

For this reason, the dates set by the NBER Committee for the beginning and the end of a recession are generally accepted as soundly based.

D.  Conclusion and the Prospects for 2023

Was the economy in a recession in the first half of 2022, as a number of  commentators have asserted?  (See, for example, this report on Fox Business, that asserted the US was in what they called a “technical recession” in the first half of 2022, or these unsurprising statements from Republican Senators Rick Scott of Florida and Rob Portman of Ohio.)

Formally, the NBER Committee has not met on this, so no such determination has yet been made.  But more fundamentally, based on the criteria the Committee uses it is highly doubtful that it will at some point decide the economy was in a recession in the first half of 2022.  The job market as well as other measures have been extremely strong.  Furthermore, even the GDP measure has been misinterpreted in the media as the Committee pays more attention to the average of the estimates for the expenditure-based GDP and the income-based GDI rather than just the former.  By this measure, the economy in fact grew in the first half of 2022 – although not by much and where future revisions in the data might change this.  But even if future data should indicate there was in fact a decline, it would certainly not be by much.

I should hasten to add that this does not mean the economy might not soon be in a recession.  Personally, I believe there is a significant possibility that the economy will be in a recession in 2023, possibly starting later in 2022.  Government spending is coming down sharply from the giant packages passed under Trump in 2020 and then continued under Biden in 2021 to provide relief from the Covid crisis; households are now spending savings that some had accumulated during the pandemic period; and the Fed is raising interest rates with the deliberate intent to slow the economy in order to reduce inflation.  I will expand on each of these in turn.

Using data from the Congressional Budget Office, total federal government spending rose by $2.1 trillion dollars in FY2020 under Trump, an increase of close to 50% from the $4.4 trillion spent in FY2019.  It rose from 21.0% of GDP in FY2019 to 31.3% of GDP in FY2020.  That was gigantic and unprecedented in the US other than during World War II.  It then stayed at roughly that level in FY2021, the first year under Biden (or rather two-thirds of a year as Biden was inaugurated on January 20 and the fiscal year starts on October 1).  In FY2021 federal government spending in fact fell as a share of GDP to 30.5% while rising in dollar value by $269 billion.  But in FY2022 it has now been reduced under Biden by $1.0 trillion – falling as a share of GDP by 7 percentage points to 23.5% of GDP.  There has not been such a fall in government spending since 1947 (as a share of GDP).

In terms of the federal government fiscal deficit, the deficit was at 4.7% of GDP in FY2019 (already substantially higher under Trump than the 2.4% of GDP it was in FY2015, as Trump increased spending while cutting taxes – mostly on the rich and on corporations).  The deficit then jumped to an unprecedented level (other than during World War II) of 15.0% of GDP in FY2020, before falling to 12.4% of GDP in FY2021 under Biden and an expected 3.9% of GDP in FY2022.  Note that this deficit in FY2022 is well less than the 4.7% of GDP in FY2019 under Trump before the Covid crisis.

This sharp cutback in federal government spending under Biden (not the story normally told by Republican politicians) would in itself be deflationary.  It has not been, however, as households as well as businesses are now spending balances many had saved and built up in 2020 and continuing into 2021.  These saving balances were built up from what they received under the various government support programs as well as due to other Covid-related programs (such as the option to suspend payments on certain debts), while spending was kept down (one did not go out to eat at restaurants as often, if at all, for example).  Note this was not the case for everyone.  Many households could only continue to barely get by – spending what they received.  But for other households, the programs led them to increase their savings balances.

The constraints on spending lifted during the course of 2021, and as accumulated savings were spent there was greater demand for goods than supply.  Prices were bid up despite the sharp cutback in government spending in FY2022.  Amplified also as a consequence of the Russian invasion of Ukraine in February 2022 that led to jumps in the prices of foods and fuels, the year-on-year increase in the CPI hit 9.1% in June 2022, before falling some to a still high 8.5% in July 2022.

The jump in the CPI – which started in mid-2021 – has led the Fed to raise interest rates.  Their aim is that the higher interest cost will lead to lower investment, which will reduce aggregate demand.  It hopes to do this without tipping the economy into a recession, but coupled with the sharp cuts in federal government spending and depletion of the excess savings that had built up during the pandemic, there is a significant danger that the Fed will not succeed in this.

It is always tricky, as interest rates are a blunt instrument for moving the economy.  Also, interest rates affect demand only with some lag that is hard to predict.  Finally, if a sharper than desired downturn does appear imminent and some boost in federal government spending becomes warranted to offset this, a Congress controlled by Republicans following the November elections would almost certainly block this.  As discussed above, one saw such dynamics during the Obama presidency following the election of a Republican-controlled Congress in November 2010.  They forced through government spending cuts in the subsequent years, despite the still weak economy following the 2008/09 collapse – the first time there were such cuts in government spending (since at least the 1970s) when unemployment was still high following a recession.  This slowed the pace of the recovery.

There could very well be a repeat of that mistake in 2023.  A recession cannot be ruled out.

Contribution to GDP Growth of the Change in Inventories: Econ 101 Again

A.  Introduction

The contribution of changes in inventories to changes in reported GDP is easily misunderstood.  One saw this in reports on the recent release (on July 28) by the Bureau of Economic Analysis (BEA) of its first estimate of GDP for the second quarter of 2022.  It estimated that GDP fell – at an annualized rate of -0.9% in the quarter – and that along with the first quarter decline in GDP (at an estimated rate of -1.6%), the US has now seen two straight quarters of falling GDP.  While there will be revisions in the coming months of the second quarter figures, as additional data become available, a fall in GDP for two straight quarters has often been used as a rule of thumb for an economy being in recession.

News reports on the figures noted also that were it not for the estimated change in inventories, GDP would have gone up rather than down.  The estimate was that GDP fell by -0.9% (at an annual rate) in the second quarter, and that the change in private inventories alone accounted for a 2.0% point reduction in GDP.  That is, if the inventory contribution had been neutral, GDP would have grown by about 1% rather than fallen by almost 1%.

But it would be wrong to attribute this to “decreases in inventories”, as some reports did.  Inventories grew strongly in the fourth quarter of 2021, with this continuing at a similarly strong pace in the first quarter of 2022 and still (although at a slower pace) in the second quarter of 2022.  How, then, could this have contributed to a reduction in GDP in 2022?

It is easy to become confused on this.  While really just a consequence of some basic arithmetic, it does require a good understanding of what GDP is and how changes in inventories are reflected in GDP.  I discussed this in a January 2012 post on this blog, but that was more than a decade ago and a revisit to the issue may be warranted.  This post will examine the problem from a different perspective from that used before.  It will start with a review of what GDP measures, and then use some simple numerical examples to show how changes in inventories affect GDP.  It will then use a series of charts, based on actual numbers from the GDP accounts in recent years, to show how changes in inventories have mattered.

A note of the data:  All the figures used come from the BEA National Income and Product Accounts (NIPA), as updated through the July 28 release.  These are often also called by many (including myself) the GDP accounts, but NIPA is the more proper term.  Also, the figures for inventories in the NIPA accounts are for private inventories only.  Inventories held by government entities are small and are not broken out separately in the accounts.  Instead, changes in such inventories are aggregated into the figures for government consumption.  While I will often refer to “inventories” in this post, the measures of those inventories are technically for private inventories only.

B.  Inventories and GDP, with Some Simple Numerical Illustrations

GDP – Gross Domestic Product – is a measure of production (product).  Yet as anyone who has ever taken an Econ 101 class knows, GDP is typically described as (and measured by) how those goods and services are used:  for Consumption plus Investment plus Government Spending plus Net Foreign Trade (Exports less Imports).  In symbols:

GDP = C + I + G + (X-M)

Where “C” is private consumption; “I” is private investment; “G” is government spending on goods or services for direct consumption or investment; and “X-M” is exports minus imports, or net foreign trade.

(Imports, M, can be thought of either as an addition to the supply of available goods or netted out from exports, X, to yield net exports.  To keep the language simple, I will treat it as being netted out from exports.)

Private investment includes investment both in new fixed assets (such as buildings or machinery and equipment) and in accumulation of inventory.  This accumulation of inventory, or net change in inventory, is key to why this equation adds up.  As noted above, GDP is product – how much is produced.  Whatever is produced can then be sold for consumption, fixed asset investment, government spending on consumption or investment, or net exports.  If whatever is produced exceeds what is sold in the period for these various purposes, then the difference will accrue as inventories.  If the amount produced falls short of what is sold, there will have to have been a drawdown of inventories for the demands to have been met.  Otherwise it would not have been possible – the goods had to come from somewhere.

The balancing item is therefore the change in inventories.  It is what allows us to go from an estimate of what is sold to an estimate (if one knows how much inventories changed by) of what was produced, i.e. to Gross Domestic Product.

How then do changes in inventories affect measured GDP?  This is best seen through a series of simple numerical examples, tracing changes in the stock of inventories over time.




Change in the Change










Start with a stock of inventories in the economy as a whole in period 0 of say 2000 (in whatever units – perhaps billions of dollars).  This stock then grows to 2200 in period 1 and 2400 in period 2.  The change in inventories in period 1 will then be 200, and that change in inventories will be one of the components making up GDP (along with private consumption, private fixed investment, and so on).  It is an investment – an investment in inventories – and thus one of the uses of whatever product was produced in the period.  It will equal the total of what was produced (GDP) less what was sold for the sum of all final demands (private consumption, private fixed Investment, government, and net foreign trade).

With the stock of inventories growing to 2400 in period 2, the change in inventories in that period will once again be 200.  Hence the contribution to GDP will once again be 200.  This is the same as what its contribution to GDP was in the previous period, and hence the higher inventories would not have been a contributor to some higher level of GDP – its contribution to GDP is the same as before.  The change in the change in the stock of inventories is zero.

But this does not mean that inventories fell in period 2.  They grew by 200.  But that was simply the same as its accumulation in the prior period, so it did not add to GDP growth.

To make a contribution to GDP growth in period 2, the addition to inventories would have had to have grown.  For example:




Change in the Change










In this example, the stock of inventories grew to 2500 in period 2.  The change in inventories was then 300, which is higher than the change in inventories of 200 in period 2 – it is 100 more.  This would be reflected in a GDP in period 2 which would be 100 higher than it would have been otherwise.

If, on the other hand, the pace of inventory accumulation slows, then inventory accumulation will subtract from GDP:




Change in the Change










In this example, inventories are still growing in period 2 – to a level of 2300.  This is 100 higher than what it was in period 2.  But the change in inventories is then only 100 – which is less than the change of 200 in period 1.  Inventories are still growing but they will add less to GDP than they had in period 2.  Hence they will subtract from whatever growth in GDP there might have been otherwise.

This is what happened in the recently released estimates for GDP growth in the second quarter of 2022.  Inventories were still growing, but they were growing at a slower pace than in the prior quarter.  In terms of annual rates (and with seasonally adjusted figures), inventories grew by $81.6 billion in the second quarter (in terms of constant 2012 dollar prices; see line 40 of Table 3 of the BEA release).  But this was less than the $188.5 billion growth in inventories in the first quarter of 2022.  In percentage point terms, that difference (a reduction of $106.8 billion) subtracted 2.0% from what GDP growth would have otherwise been in the second quarter (see line 40 of Table 2 of the BEA release).  With the changes in the other components of GDP, the end result was that estimated GDP fell by 0.9% in the quarter.  Thus one can attribute the fall in GDP in the quarter to what happened to inventories, but not because inventories fell.  It was because they did not grow as fast as they had in the previous quarter.

C.  Changes in Inventories in the Data

Based on this, it is of interest to see how inventories have in fact changed quarter to quarter in recent years.  These changes, and especially the changes in the changes, are volatile.  They can make a big difference in the quarter-to-quarter changes in GDP.  Over time, however, they will even out, as there is some desired level of inventories in relation to their sales and producers will target their purchases to levels to try to reach that desired level.

Start with the chart at the top of this post.  It shows the stock of private inventories by quarter going back to 1998.  The figures are in constant 2012 dollars so that inflation is not a factor (and more precisely using what are called “chained” dollars where the weights used to compute the overall indices are based on prior period shares of each of the goods – so the weights shift over time as these shares shift).

Stocks generally move up over time as the economy grows, although there have been reductions in periods when the economy was in recession or otherwise disrupted.  Thus one sees a fall in 2001, due to the recession in the first year of the Bush II administration, an especially sharp fall in 2008 with the onset of the economic and financial collapse in the last year of the Bush II administration with this then carrying over into 2009, and then a fall again in 2020 due to the Covid lockdowns.  The trough in the most recent downturn was reached in the third quarter of 2021, following which the stock of inventories grew rapidly.  They are still, however, slightly below the level reached in mid-2019 even though GDP is higher now than what it was then.

One starts with the stocks, but as was discussed above, the contribution to GDP comes from the accumulation of inventories – the change in the stocks.  These changes, based on the figures underlying the chart at the top of this post, have been:

There is considerable quarter-to-quarter volatility.  Note that the figures here are expressed in terms of annual rates.  That is, they are each four times what the actual change was (in dollar terms) in the given quarter.  One sees that the change in the fourth quarter of 2021 was quite high – higher than in any other quarter of this 24-year period – and was still almost as high in the first quarter of 2022.  The increase was then less in the second quarter of 2022, but was still a substantial increase (of $81.6 billion at an annual rate) in the quarter.

The changes in inventories are a component of GDP, but the contribution to the growth in GDP comes from the changes in the change in inventories.  These are easily computed as well by simple subtraction, and were:

These are now very highly volatile, and one sees especially sharp fluctuations in the last couple of years.  With all the disruptions of the lockdowns, the subsequent supply chain disruptions, and the very strong recovery of the economy in 2021 (with GDP growing faster than in any year in almost four decades, and private consumption growing faster than in any year since 1946!), it has been difficult to manage production to meet expected demands and allow for some desired target level of inventories.

This had a substantial impact on the quarter-to-quarter changes in GDP, both positive and negative.  Focussing on the recent quarters, the changes in inventories were a $193.2 billion increase in the fourth quarter of 2021, and as noted before, a further $188.5 billion increase in the first quarter of 2022 and a further although smaller increase of $81.6 billion in the second quarter of 2022.  These were the changes in inventories.  But the changes in the changes, which is what will add to or subtract from GDP growth, were a very high $260.0 billion in the fourth quarter of 2021, and then a fall of $4.7 billion in the first quarter of 2022.  This reduction in the first quarter of 2022 came despite inventories increasing in that quarter by close to a record high level.  But they followed a quarter where inventories rose by a bit more, so the change in the change was small and indeed a bit negative.

In the second quarter of 2022 inventories again rose – by $81.6 billion.  But following the close to record high growth in the first quarter of 2022, its contribution to the growth in GDP in the quarter was substantially negative.  The $81.6 billion increase in inventories in the second quarter was $106.9 billion less than the increase of $188.5 billion in the first quarter.  And it is this $106.9 billion which is a contribution to (or in this case a subtraction from) what GDP growth would have been in the quarter.

Finally, one can show this also in the possibly more helpful units of the percentage point contribution to the growth in GDP:

Although in different units, the chart here mirrors closely the preceding one, as one would expect if one has been doing the calculations correctly.  The only difference, in principle, is that with GDP growth over time, the dollar values of the quarter-to-quarter changes will look larger when expressed as a share of GDP in the earlier years of the period.

There are, however, some minor differences deriving from the nature of the data used.  The chart here was drawn directly from the figures presented in the BEA NIPA accounts for the percentage point contributions to GDP growth from changes in inventories.  One can also calculate it by taking the quarterly changes in the change in constant dollar terms (from the preceding chart, in red), dividing it by the previous quarter’s GDP (as one is looking at growth over the preceding quarter), and then annualizing it by taking one plus the ratio to the fourth power.  I did that, and the curve lies very close to on top of the curve shown here (in orange).

But not quite, due in part to rounding errors that compound when one is taking the changes and then the changes in the changes.  In addition, inventories by their nature are highly heterogeneous, with some going up and some down in any given period even though there is some bottom line total on whether the aggregate rose or fell.  This makes working with price indices tricky.  The BEA figures are based on far more disaggregated calculations than the ones they present in the NIPA accounts, and their underlying data also have more significant digits than what they show in the tables they report.

D.  Inventories to Sales, and Near Term Prospects

What will happen to inventories now?  Given how important changes in inventories are to the quarter-to-quarter figures on GDP growth, economists have long tried to develop some system to predict how they will change (as have Wall Street analysts, where success in this could make some of them very rich).  But they have all failed (at least to my knowledge).

One statistic that many focus on, quite logically, is the ratio of inventory to sales:

The figures here were computed from data reported in the BEA NIPA Accounts, Table 5.8.6B, where inventories include all private inventories while sales are of goods (including newly built structures) sold by domestic businesses.  Inventories are by nature of goods only, and hence one should leave out services (as an increasing share of services in GDP would, on its own, lead to a fall in the ratio).  Sales of newly built structures are included as one has inventories of building materials.  The figures on the sale of goods by domestic businesses are provided by the BEA.  Note that “sales” here are expressed on a monthly basis.  Hence the ratio is of inventories in terms of months of sales.

As one sees in the chart, the ratio of inventory to sales has been coming down over time.  This is consistent with all the literature advising on tighter inventory management.  There was then an unusually sharp decline in 2020 – a consequence of the Covid lockdowns – that bottomed out in the second quarter of 2021 (as a share of sales) and has since grown strongly.  But the ratio is still below where it was prior to the pre-Covid trend, although how much below depends on how one would draw the trend line pre-Covid.

Where will it go from here?  While important to what will happen to the quarter-to-quarter figures for GDP growth, as discussed above, I doubt that anyone has a good forecast of what that will be.  While there might well be room for the inventory to sales ratio to rise from where it is now, keep in mind that the ratio can rise not only by adding to inventories but also by sales going down.  And while GDP growth was exceptionally strong in 2021, it has been weak so far this year (indeed negative) and that weakness might well worsen.  Personally, while I do not see that the economy is in recession now (employment growth has been strong, with 2.7 million net new jobs in the first half of 2022, and the unemployment rate has been just 3.6% for several months now), the likelihood of a recession in 2023 is, I would say, quite high.

There also have been recent announcements by major retailers that the inventories they are currently holding are well in excess of what they want, and that they will take exceptional measures to try to bring them down.  Target announced a plan to do so in June (with a warning it will squeeze their near-term profits), Walmart announced in July they had similar issues (and that it would slash prices to move that inventory), and other retailers have announced similar problems.  If this is indeed a general issue, then those efforts to bring down inventories in themselves will act as a strong drag on the economy, making a recession even more likely.  And as was discussed above, the stock of inventories does not need to fall in absolute terms to cut GDP growth – a change that is less than what the change had been in the prior period will subtract from GDP growth, even though the inventories may still be growing in absolute terms.

Firms such as Target and Walmart employ many highly trained professionals to manage their inventories.  Yet even they find it difficult to get their inventories to come out where they want them to be.  If they and others now begin a concerted effort to bring down their inventory levels in the coming months, the impact on GDP in the rest of this year could be severe.