The Long View and the Discount Rate

Volume VII — Planetary and Cosmic

Nine movements, one horizon.

THE PLATE

A watercolour of open fields running to distant hills, wildflowers in the grass. — Plate VII.08The Marker Field.These are not headstones. They are a message to somebody ten thousand years from now, written in geometry because no language is expected to survive. Somebody costed them, approved them and signed for them — which means somebody put a number on the year twelve thousand.

THE LETTER

Chapter III.05 took the discount rate apart: the three Ramsey terms, which two of them are ethics rather than measurement, and the published declining schedules that three finance ministries already stand behind. Chapter IV.09 took that instrument and pointed it at concrete, and found that a rate does not mainly shrink the benefit of lasting — it shrinks the cost of not lasting, and by more.

Both of those chapters used the machinery. This one goes to the horizon where the machinery genuinely breaks, and says so plainly, because you will get there and it is better to arrive expecting it.

Three things happen out past a century and they are different in kind. The first is a mathematical fact and it is good news: once you admit that you do not know what the discount rate will be, the correct certainty-equivalent rate falls over time toward the lowest rate you consider possible. Not because anyone decided the future deserves more weight. Because of what averaging does. That is Weitzman's and Gollier's result and we will compute it rather than describe it.

The second is that the ethics stop being decoration. Out past three generations, the people you are discounting are not merely unknown — under any policy that touches the timing of anybody's life, they are different people, and Parfit showed that this dissolves the ordinary language of harm. It does not dissolve the obligation. Knowing exactly what it does and does not license is the difference between a serious long-horizon argument and a pious one.

The third is that a declining schedule is time-inconsistent. A government thirty years from now will want to re-optimise, and — this is the part nobody says out loud — it will be right to, by its own lights, applying the same theorem from its own present. We will measure the size of that gap and then name the six commitment devices that survive it, with what each one costs.

And then the part that pays for the chapter: what has actually been committed across generations and held. Not proposals. Funded balance sheets with published rules, and a regulatory standard written for ten thousand years.

— The Editors

DISCOVERY

What is already working

The striking fact about multi-generational commitment is that it is not aspirational. It exists, it is funded, and the rules are published.

Norway's Government Pension Fund Global. The oil fund holds on the order of nineteen trillion kroner. What matters here is not the size but the rule bolted to it: the handlingsregelen, adopted in 2001, caps the structural non-oil deficit at the fund's expected real return, originally four percent and lowered to three in 2017. That single percentage point is the entire intergenerational transfer, and we will price it in the next movement. The fund has returned about four percent a year real, net of costs, since 1998. The rule is a political convention rather than a statute, and it has held for two decades across changes of government, which is a stronger fact about institutions than any law would be.

The Alaska Permanent Fund. In 1976 the voters of Alaska amended their own constitution — Article IX, section 15 — to require that at least twenty-five percent of mineral royalties be deposited into a fund whose principal may not be spent. The fund stands at around eighty-one billion dollars, has paid a dividend to every resident since 1982, and since 2018 draws under a percent-of-market-value rule of five percent of the five-year average. A generation of Alaskans wrote a constraint on themselves into the hardest document they had, and it has bound every legislature since, including several that badly wanted out.

Germany's KENFO. In 2017 the German nuclear operators transferred twenty-four point one billion euros into a public foundation and the state assumed the disposal liability in exchange. The entity that created the waste no longer holds the duty, because a utility is not an institution that can be relied on to exist a century from now and a sovereign fund with a published mandate has a better chance.

The United States' decommissioning trusts. Under 10 CFR 50.75 every reactor licensee must hold externally segregated funds against decommissioning and report its funding status against a published formula minimum every two years. The minimum has a base year — January 1986 — and two base figures: one hundred and five million dollars for a large pressurised water reactor, one hundred and thirty-five for a boiling water reactor, escalated by formula. The commitment device is not the fund. It is the biennial public report against a number nobody gets to choose.

France. The law of 28 June 2006 requires operators to hold dedicated, segregated assets covering their long-term nuclear obligations, with coverage tested and reported. The United Kingdom. The Nuclear Liabilities Fund holds assets against the AGR fleet's decommissioning while the Nuclear Decommissioning Authority carries a provision of the order of one hundred and thirty billion pounds undiscounted, spread across roughly one hundred and twenty years. Finland. Posiva's repository at Onkalo is under construction, funded by a statutory fund that must be kept fully provisioned, and it is the first deep geological disposal facility for spent fuel anywhere to reach that stage.

And the regulatory object that proves the point. The Environmental Protection Agency's standard for Yucca Mountain, 40 CFR Part 197, originally required compliance for ten thousand years. The National Academies had already reported in 1995 that there was no scientific basis for that cut-off, because peak dose arrives hundreds of thousands of years later. The D.C. Circuit vacated the limit in NRDC v. EPA in 2004, and the 2008 final rule extended the standard to one million years — fifteen millirem a year to year ten thousand, one hundred millirem median thereafter.

Read what that is. A federal agency, under judicial instruction, wrote a binding performance standard covering a period forty times longer than recorded history. It did not do so by choosing a very low discount rate. It did so by not using one. Hold that; the Arithmetic will show why it had no alternative.

THE ARITHMETIC

What works, what does not, and where the line sits

First: the certainty-equivalent rate falls, and it is not a preference.

Suppose the true long-run rate is fixed but unknown. Take seven candidate rates from one to seven percent, equally likely — a mean of exactly four percent. You must average discount factors, never rates, because the factor is what a pound is worth and the rate is only its logarithm.

  t      E[e^-rt]      spot R(t)   forward f(t)   weight on the 1% branch
      0   0.999996       4.0000%      4.0000%            14.29%
     10   0.683805       3.8008%      3.6033%            18.90%
     50   0.213564       3.0876%      2.3235%            40.57%
    100   0.083064       2.4881%      1.5756%            63.27%
    200   0.022360       1.9002%      1.1565%            86.47%
    300   0.007485       1.6316%      1.0524%            95.02%
    400   0.002665       1.4819%      1.0187%            98.17%
   1000   0.000006       1.1946%      1.0000%           100.00%

The mechanism is in the last column and it is entirely mechanical. The forward rate is the average of the candidate rates reweighted by which branches have any weight left. By year 100 the one-percent branch already carries 63.3 percent of the surviving mass; by year 400, 98.2 percent; by year 1,000 the seven-percent branch contributes e^-70, which is about four parts in ten thousand billion billion billion, and the average is simply the lowest rate.

The far-distant future is discounted at the lowest rate you think possible. That is Weitzman's 1998 title and it is a theorem, not a position. Gollier and Weitzman settled the remaining objection in 2010: when the representative agent re-optimises against the uncertainty rather than passively holding, the effective rate still declines to its lowest possible value.

Second, and this is the cut: widening the uncertainty makes the far future worth more. Hold the mean at exactly four percent and change only the spread.

  distribution           mean      sd     R(300)    PV of GBP 1m at y300
  certain, 4%            4.00%   0.00%    4.0000%              6.14
  narrow, 3-4-5%         4.00%   0.82%    3.3492%             43.29
  wide,   2-4-6%         4.00%   1.63%    2.3654%            828.30
  wider,  1-4-7%         4.00%   2.45%    1.3662%         16,597.74
  widest, 0.5-4-7.5%     4.00%   2.86%    0.8662%         74,378.77

Same mean, every row. Going from certainty to a spread of half a percent to seven and a half multiplies the value of a year-300 benefit by 12,106 times.

Stop on that, because it inverts the thing almost everyone believes. The standard move against long-horizon commitment is we cannot possibly know anything about the year 2300, so it is unserious to plan for it. The arithmetic says the opposite. Uncertainty is not a reason to discount the far future more. It is, by itself, the reason to discount it less — and the deeper the ignorance, the heavier the future's weight. Ignorance about the rate is an argument for the long view, and it has a number attached.

Third: where the closed form breaks, stated because it matters. If rates are normally distributed, E[e^-rt] = exp(-mu·t + sigma²t²/2) exactly, so the spot rate is R(t) = mu - sigma²t/2 — linear in t. With a four percent mean and a two-point standard deviation it falls a full percentage point every fifty years, reaches zero at 2mu/sigma² = 200 years, and goes negative after. That is not a discovery about the future; it is an artefact of a distribution with no floor. The theorem's limit is the lowest rate in the support, so a model without a floor has no answer. Any declining schedule is only as defensible as the lower bound somebody chose, and choosing that bound is not arithmetic.

Fourth: growth risk is not where the decline comes from. Gollier's extended Ramsey rule adds a precautionary term for uncertain growth: `r(t) = delta + eta·ĝ

(eta²/2)·V(t)/t, where V(t)` is the variance of cumulative log growth. With

delta 0.5 percent, eta 2, growth 1.5 percent and a two percent standard deviation, the no-risk rate is 3.50 percent, and the precautionary term depends almost entirely on how persistent growth shocks are:

  persistence rho    precautionary term    r as t -> infinity
        0.0                0.080 pp              3.4200%
        0.3                0.149 pp              3.3514%
        0.7                0.453 pp              3.0467%
        0.9                1.520 pp              1.9800%

Rate uncertainty moves the rate by 3.00 percentage points. Growth uncertainty at a persistence of 0.7 moves it by 0.45 — a factor of 6.6. The declining schedule is a fact about our ignorance of the rate, not about the riskiness of growth, and a paper that defends a declining schedule on precautionary grounds has defended it on the weaker of the two arguments.

Fifth: the horizon where the instrument stops being an instrument.

  GBP 1 arriving in year 10,000    at 1.0%   =  10^-43.2 pounds
  GBP 1 arriving in year 300,000   at 1.0%   =  10^-1,296 pounds
  GBP 1 arriving in year 1,000,000 at 1.0%   =  10^-4,321 pounds

One percent is the Green Book's floor — the lowest rate in the best-constructed public schedule in the world, applied to everything beyond year 300. It still prices a million-year dose at ten to the minus four thousand pounds. There is no rate low enough, no fund large enough and no willingness-to-pay survey fine enough. Past a few thousand years, present value is not a bad answer. It is not an answer. The EPA did not choose a low rate for Yucca Mountain. It wrote a dose ceiling in millirem, with no price on it, because a constraint is the only object that survives the horizon.

Sixth: the ethics, and they are arithmetic too.

Three problems sit inside every long-horizon calculation and are almost never declared.

Ramsey's own view of pure time preference. The man who wrote the equation called discounting future utilities "ethically indefensible" and said it "arises merely from the weakness of the imagination." Sidgwick had said the same in 1874; Pigou called it a defective telescopic faculty; Harrod called it a polite expression for rapacity. Ramsey set delta to zero — and then needed a device, because with delta = 0 an infinite horizon has no finite total to maximise. A flow of one per year for ten thousand years sums to 10,000 at zero and to 200.00 at half a percent, which is 1/0.005. Ramsey's answer was the Bliss point: maximise the integral of (B - u), which converges. Koopmans in 1960 and Diamond in 1965 later proved there is no complete, continuous, anonymous and Pareto ordering of infinite streams at all. Zero pure time preference is the right ethics and it does not, on its own, give you a maximum. That is the honest position: the defensible value of delta is the hazard of extinction, and the technical work of getting a well-defined optimum has to be done elsewhere.

Parfit's non-identity problem, computed. Any policy that alters the timing of anybody's life alters who is conceived. A person exists only if every ancestral conception happened at the moment it did, and going back k generations there are 2^k - 1 of those. Let q be the per-generation, per-lineage probability that a policy shifts one:

  generations k   ancestors 2^k-1    q = 0.001     q = 0.010     q = 0.050
        3                   7          0.9930        0.9321        0.6983
        7                 127          0.8807        0.2790        0.0015
       10               1,023          0.3593      3.43e-05      1.63e-23
       12               4,095          0.0166      1.34e-18      6.00e-92

At a one-percent perturbation the seventh generation is 27.9 percent likely to contain the same people, and by the twelfth it is one part in a million million million. So a person in 2300 cannot say they are worse off than they would have been. They would not have been.

What that licenses is narrower than either side usually claims. It defeats a claim of harm to an individual. It leaves every claim about the quality of the world standing — Parfit's own same-number quality claim: if the same number of people will live either way, it is worse if those who live are worse off than those who would have lived. Read what that does to the instrument choice. A net present value of individual harms is exactly the object the non-identity problem dissolves. A dose ceiling, a stock level, a boundary — a statement about the state of the world — is exactly the object it leaves untouched. The ethics and the arithmetic arrive at the same instrument from opposite directions.

The population term, which every long-horizon calculation quietly contains. Under a total criterion the planner sums over people, so the rate on per-head consumption is delta - n + eta·g, not delta + eta·g. On UN medium projections world population runs from 8.2 billion in 2024 to about 10.2 billion in 2100 — n of 0.287 percent a year:

  2100 population    n per year    r = 3.5% - n    PV of GBP 1m at year 76
  low    ~7.0 bn       -0.2082%        3.7082%                  62,836
  average criterion     0.0000%        3.5000%                  73,204
  medium  10.2 bn       0.2872%        3.2128%                  90,415
  high    12.0 bn       0.5010%        2.9990%                 105,851

The medium variant raises the future's weight by 1.24 times. The low variant lowers it by 0.86, because n is then negative and a total criterion discounts the future more heavily. The criterion most often reached for to defend future generations changes sign when population peaks — which, on the UN's own medium variant, it does in the 2080s. That is not an argument against the total view. It is an argument for declaring which view you are using, on the page, beside the rate.

Seventh — the honest negative. A declining schedule is time-inconsistent, and your successor is not cheating.

Take the Green Book table and a benefit of one million pounds landing in year

Today it is worth £6,207. Now stand at year 30 and look again.

  standing at year 30, with 170 years still to run
    the original plan's own continuation value     GBP 17,423
    the successor, re-applying the same table       GBP 11,244
    the successor values it at                          64.5%
    the plan over-states by                             1.550x

  standing at year 75, with 125 still to run          41.6%   /   2.404x

Nobody has done anything improper. The successor applied the published table from its own present, which is exactly what the table instructs. The gap exists because the schedule is indexed by years-from-now, and years-from-now re-bases with every government. Weitzman's own schedule is not like this: it is indexed to calendar date and to what has been learned, and once you have lived through thirty years you genuinely have information about which branch you are on. That schedule is time-consistent. No finance ministry publishes it. What they publish is the horizon-indexed table, which is a permanent standing incentive for every successor to reprice your commitment down by a third and be right about it.

That is the real limit on the long view, and it is not solved by finding a lower rate. It is solved by taking the decision out of the discounting frame, which is what the surviving institutions did.

DREAM

What becomes ordinary

In the organisation that has absorbed this, the long-horizon obligation has a balance sheet and not an intention.

Every duty that outlives the people who incurred it — decommissioning, closure, aftercare, the pension promise, the land covenant, the tailings dam — sits in a segregated fund with a named external trustee, and the fund publishes one number on its first page: the real return it is implicitly asserting. Not the funding ratio. Funding ratios are read aloud in board meetings as though they were facts, and they are discount rates wearing a percentage sign. The implied required return is the honest form of the same quantity, and it can be compared to something — the fund's own realised return over the last twenty years — which is what makes it usable.

The appraisal pack carries a declining schedule, and beside it, on the same page, carries its calendar form: the forward rate this plan is committing successors to for the half-century after 2075, stated as a number, so that a successor who departs from it must do so visibly. The gap between the horizon schedule and the calendar schedule is disclosed rather than discovered.

And past a certain horizon the pack stops quoting present values at all. Where the duty runs to thousands of years, the paper carries a standard — a concentration, a dose, a stock level, a state of the world — with a compliance test and a monitoring interval, and the finance function treats meeting that standard at least cost as the optimisation, which it is. Nobody experiences this as a defeat for economics. It is economics behaving correctly at a horizon where prices have stopped carrying information, and choosing the instrument that still does.

The people who set these things know what they have signed. There is a document, it names the duty, it names the fund, it names the trustee, and it names what would have to happen for it to be undone — which is a specified and difficult thing rather than a quiet absence of renewal.

DESIGN

The structure that gets there

Six commitment devices survive a successor's re-optimisation. Each one costs something, and the cost is the reason to choose between them rather than collect them.

One — constitutional entrenchment. Alaska, 1976: the deposit requirement and the inviolability of principal are in the constitution and can be changed only by a vote of the people. What it costs: everything, in a crisis. Alaska ran multi-billion-dollar deficits with tens of billions of untouchable principal beside it, and the adjustment fell on services and on the dividend instead. An entrenched rule is a rule you cannot use when you most want to, which is the point and also the price.

Two — a published formula plus an independent scorekeeper. Norway's fiscal rule is not law. It is an arithmetic convention, published, forecast by an independent body, and breachable. What it costs: it holds by reputation, and reputation is an asset the first defector destroys. It has held for two decades, which is evidence, not a guarantee.

Three — an irrevocable external trust with a use restriction. The NRC's segregated decommissioning funds, KENFO, France's dedicated assets. Money leaves the balance sheet, a trustee holds it, and it may be spent on one thing. What it costs: you have not escaped the discount rate. You have relocated it into an assumed return, which is at least a number somebody must publish and be measured against annually. That relocation is the whole value of the device.

Four — a counterparty with money at stake. An insurer, a bondholder, a creditor whose covenant is breached if the fund falls short. What it costs: it prices into your cost of capital, and the counterparty is very likely shorter- lived than the duty.

Five — physical irreversibility. A closed repository. A conservation easement running with the land. A dismantled structure. What it costs: the option value of changing your mind, which IV.09 priced and found is usually smaller than people fear and is never zero. It also only works where the commitment is physical.

Six — replace the rate with a standard. Express the duty as a state of the world — a dose ceiling, a stock level, a boundary — with a compliance test, and optimise cost subject to it rather than trading it off. What it costs: it is unpriced, so it cannot be balanced against anything, which makes it both extremely robust and occasionally very expensive. It is what the EPA did at Yucca Mountain and it is the only device on this list that works at ten thousand years.

Sequence. Segregate and appoint a trustee first, because it is a week's legal work and it converts an intention into an asset. Publish the implied required return second, because it costs nothing and it is the number that makes the fund arguable. Adopt the calendar-form schedule third. Move to a standard only where the horizon genuinely defeats the rate — which is rarer than enthusiasts think and commoner than treasuries think.

Governance. The duty, the fund, the trustee, the assumed return, the review date: five items, one page of policy, signed, reviewed annually. Not a spreadsheet template. A number in a template is owned by whoever last edited it.

DESTINY

How it holds when nobody is pushing

It holds when the implied required return is published on the face of the accounts, because an assumption that must be disclosed is an assumption that must be defended, and a fund asserting 7.2 percent real forever cannot be defended for long in daylight.

It holds when the reporting is biennial, public and formulaic, which is the one thing the NRC regime gets unambiguously right. A licensee cannot quietly drift; it must file against a number it did not choose.

It holds when the duty and the money are held by different institutions. The utility that created German waste no longer carries the liability; a foundation does. Separating the party that benefits from the party that owes is the oldest device in trust law and it is the one most often skipped.

And here is how it fails. It fails exactly where the United States failed. The Nuclear Waste Fund collected one mill per kilowatt hour under the 1982 Act — about seven hundred and eighty million dollars a year at recent generation — faithfully, for three decades, and the fee was set to zero in 2014 after NARUC v. DOE, and the repository was still not built. Funding a duty and discharging it are two different commitments, and only one of them was instrumented. A fund with no delivery covenant produces a very large pot of money and an undischarged obligation, which is a worse position than either alone, because it looks solved.

It fails, more quietly, when the schedule is horizon-indexed and the successor does the arithmetic correctly. That is the 64.5 percent above, and the defence is the calendar form, published, so the reprice is a visible act rather than a routine recalculation.

DELIGHT

What it feels like

There is a particular quiet in a room where a horizon has been named out loud and nobody flinched. Somebody asks how long, and instead of the usual shrug toward the long term, the answer is a century, or twelve hundred years, or a million, and the paper on the table has that number on it, and the conversation carries on. The relief is physical. A horizon you can say is a horizon you can work in.

Then there is the pleasure of the inversion landing. You have spent years being told that the far future is unknowable and therefore not your business, and one afternoon there is a line of arithmetic showing that the unknowability is the argument on your side — that the wider the spread of futures you are willing to admit, the more weight the far one carries, by a factor of twelve thousand. That is not consolation. It is a better argument than the one you had, and it was sitting inside the objection.

And there is something companionable about the markers. Somebody sat in an office and designed a field of stones to say this place is not a place of honour to a person who will not share a language with us. They costed it. They approved it. It is the least cynical act in the whole of public finance, and it was done by engineers and lawyers doing ordinary jobs on an ordinary Tuesday.

OPERATIONALIZE THIS

At the level of finance

The instrument: a segregated long-horizon obligation trust with a published implied-return covenant and a restatement gap.

Not a provision on the balance sheet, which is an estimate the issuer controls. Not a green bond, which prices off credit. This is narrower: a funded trust whose defining disclosure is the real return it is asserting, tested annually against realised return, with a mandatory top-up when the gap persists.

The mechanics.

Structure. An irrevocable trust with an external corporate trustee and a single permitted use. Assets leave the sponsor's balance sheet. Precedent: 10 CFR 50.75 external sinking funds; KENFO; French dedicated assets.
The liability schedule. The duty stated undiscounted, by calendar year, to its end. Not a single number. The UK case is instructive: roughly £130bn over 120 years, a level £1.083bn a year.
The covenant, and it is the whole instrument. The trust publishes the real return implied by its own assets against that schedule. Invert the funding ratio and read the rate:

  assets, GBP bn      implied real return, forever
        10.00                   10.83%
        15.00                    7.22%
        20.00                    5.41%
        26.84                    4.00%
        40.00                    2.58%

The restatement gap. Implied required return minus trailing twenty-year realised real return. Positive and persistent for three years, the sponsor tops up to close it. This is the covenant a credit committee will actually price, because it is objective and annual.
Term. The life of the duty, not the life of the asset, the lease or the licence.
Security. The trust corpus and the top-up covenant, in that order.

The balance-sheet treatment. Under IAS 37 the provision is measured at the present value of the expenditures expected to settle the obligation, and the rate is disclosed. Do the reverse disclosure as well: state the rate the funding asserts, beside the rate the provision uses. Where they differ — and they almost always do — the difference is the honest measure of how funded you are. Under IFRS S2 the long-horizon narrative is now disclosable, so this arithmetic has an external audience for the first time.

The counterparty. An independent corporate trustee, and the auditor as verifier of the implied return. Where the duty is regulated, the regulator's published formula minimum is a free, defensible floor that nobody in the organisation gets to argue about — which is exactly its value.

The number that decides it. One line, on the front page:

   implied required real return (assets vs undiscounted schedule)
   -------------------------------------------------------------  <=  1.0
        trailing 20-year realised real return of the fund

A ratio at or below one means the duty is funded on evidence. Above one, the organisation is not funded — it is forecasting, and the forecast is stated nowhere on the face of the accounts. The UK-scale example: £15bn against £130bn over 120 years is not "twelve percent funded". It asserts 7.22 percent real, forever, against a fund realising four. The required pot at four percent is £26.84bn. That is the sentence the paper should open with.

The first ninety days.

Day	Action	Artifact
1–15	List every duty outliving its incurring cohort; state each undiscounted, by calendar year	The liability schedule
16–30	Compute the implied required return for each existing pot	The implied-return page
31–45	Pull the fund's trailing twenty-year realised real return	The restatement gap, first cut
46–60	Draft the trust: trustee, permitted use, top-up covenant	Trust deed, first draft
61–75	Publish the schedule in calendar form beside the horizon form	Treasury policy amendment
76–90	Board adopts the covenant and the review date	Signed board minute

APPRECIATIVE QUESTIONS

Twelve, for a room

Discovery — what is already working

Which commitment made by people who worked here before us is still holding — and what specifically made it survive them?
Where do we already hold money against a duty that nobody alive here will see discharged, and who set the amount?
When has this organisation chosen a standard rather than a price for something — a limit, a floor, a ceiling — and what did that make possible?

Dream — what becomes possible

If every long-lived duty we carry published the real return it is assuming, what would become arguable next year that is only assertable now?
Imagine a paper in front of this board with a horizon of two hundred years on its first page and nobody in the room finds that strange. What is on page two?
If we could hand our successors one thing that made their job easier rather than harder, what would it be?

Design — what we build

What is the smallest duty we could put into a properly segregated trust this quarter, and whose signature does that need?
Which of our commitments would we be content for a successor to reprice, and which would we not? What distinguishes the two lists?
Where is a state-of-the-world standard the right instrument for us, and how would we know we had reached for it too early?

Destiny — how it holds

What would have to be true for the implied-return disclosure still to be on the front page when everyone in this room has moved on?
If a future board wanted to undo this, what is the most likely route, and what would make that route visible rather than quiet?
Which of our funded duties has money but no delivery covenant — and who would notice first if the money were there and the thing never happened?

WORKS CITED

Sidgwick, H. (1874). The Methods of Ethics. Macmillan.

Pigou, A. C. (1920). The Economics of Welfare. Macmillan.

Ramsey, F. P. (1928). "A Mathematical Theory of Saving." The Economic Journal, 38(152), 543–559.

Harrod, R. F. (1948). Towards a Dynamic Economics. Macmillan.

Strotz, R. H. (1955–56). "Myopia and Inconsistency in Dynamic Utility Maximization." The Review of Economic Studies, 23(3), 165–180.

Koopmans, T. C. (1960). "Stationary Ordinal Utility and Impatience." Econometrica, 28(2), 287–309.

Diamond, P. A. (1965). "The Evaluation of Infinite Utility Streams." Econometrica, 33(1), 170–177.

Parfit, D. (1984). Reasons and Persons. Oxford University Press. (Part IV, especially chs. 16–19: the non-identity problem, the Repugnant Conclusion, and Theory X.)

Nuclear Waste Policy Act of 1982, Pub. L. 97-425, and 10 CFR 50.75, Reporting and recordkeeping for decommissioning planning. United States.

National Research Council (1995). Technical Bases for Yucca Mountain Standards. National Academy Press.

Broome, J. (1994). "Discounting the Future." Philosophy & Public Affairs, 23(2), 128–156.

Weitzman, M. L. (1998). "Why the Far-Distant Future Should Be Discounted at Its Lowest Possible Rate." Journal of Environmental Economics and Management, 36(3), 201–208.

Weitzman, M. L. (2001). "Gamma Discounting." American Economic Review, 91(1), 260–271.

Newell, R. G. and Pizer, W. A. (2003). "Discounting the Distant Future: How Much Do Uncertain Rates Increase Valuations?" Journal of Environmental Economics and Management, 46(1), 52–71.

Natural Resources Defense Council v. Environmental Protection Agency, 373 F.3d 1251 (D.C. Cir. 2004).

Broome, J. (2004). Weighing Lives. Oxford University Press.

Environmental Protection Agency (2008). Public Health and Environmental Radiation Protection Standards for Yucca Mountain, Nevada; Final Rule. 40 CFR Part 197, 73 Fed. Reg. 61256.

Weitzman, M. L. (2009). "On Modeling and Interpreting the Economics of Catastrophic Climate Change." The Review of Economics and Statistics, 91(1), 1–19.

Gollier, C. and Weitzman, M. L. (2010). "How Should the Distant Future Be Discounted When Discount Rates Are Uncertain?" Economics Letters, 107(3), 350–353.

Gollier, C. (2012). Pricing the Planet's Future: The Economics of Discounting in an Uncertain World. Princeton University Press.

Arrow, K. J., Cropper, M. L., Gollier, C., Groom, B., Heal, G. M., Newell, R. G., Nordhaus, W. D., Pindyck, R. S., Pizer, W. A., Portney, P. R., Sterner, T., Tol, R. S. J. and Weitzman, M. L. (2013). "Determining Benefits and Costs for Future Generations." Science, 341(6144), 349–350.

National Association of Regulatory Utility Commissioners v. United States Department of Energy, 736 F.3d 517 (D.C. Cir. 2013).

Greaves, H. (2017). "Discounting for Public Policy: A Survey." Economics and Philosophy, 33(3), 391–439.

Parfit, D. (2017). "Future People, the Non-Identity Problem, and Person-Affecting Principles." Philosophy & Public Affairs, 45(2), 118–157.

Drupp, M. A., Freeman, M. C., Groom, B. and Nesje, F. (2018). "Discounting Disentangled." American Economic Journal: Economic Policy, 10(4), 109–134.

Norges Bank Investment Management (2024). Government Pension Fund Global: Annual Report. Oslo.

Alaska Permanent Fund Corporation (2024). Annual Report. Juneau. (And Alaska Constitution, Article IX, section 15; Senate Bill 26, 2018.)

HM Treasury (2022). The Green Book: Central Government Guidance on Appraisal and Evaluation. Table 6.1, social time preference rates.

Nuclear Decommissioning Authority (2024). Annual Report and Accounts. United Kingdom.

United Nations, Department of Economic and Social Affairs (2024). World Population Prospects 2024.

Note on figures. Every number above is computed in lib/verify/VII_08.py and printed there with its inputs, its intermediate terms, and the form the page shows it in. Figures read from a source rather than derived are labelled SOURCED; figures chosen to make a comparison visible are labelled ILLUSTRATIVE or ASSUMED. Chapters III.05 and IV.09 carry the machinery this chapter takes to its limit.