The Models

The mathematics behind every Packet.

Quantitative trading firms price risk with mathematics: probabilities, simulations, the value of every piece of information. We do the same for one building, before you buy it. Every model on this page is run on the example building, and every one ends in pounds.

Scroll to run the building

Run it ten thousand times

Every future the building could have.

Five years, a quarter at a time. The tenant stays or goes, markets fall or climb, the roof is paid for. Each line is one future, and where it ends is what you got back for every pound you put in.

The line

Below the line, you lose money.

Pay the asking price, and 26% of the futures end below it: about one in four.

Pay what the evidence supports

The same building, at a better price.

At the opening price of £7,450,000, 10% end below the line: one in ten. Nothing about the building changed. Only the price.

One number

From ten thousand futures to one decision.

The spread becomes three figures: what it is worth, where to open, where to stop. This is one model. Below are some of the others, each run on the same building.

§ 00Why models

Most buyers are toldwhat a building is worth.You will know how sure to be.

A valuation gives you one number and asks you to trust it. A model gives you every number the building could be worth, how likely each one is, and what would change it.

That is how the firms that trade markets for a living price risk. Here it is turned on a single warehouse. Every model below is shown in three layers: the question in plain words, the answer in pounds, and the machinery underneath, for anyone who wants to see how deep it goes.

IWhat it is worthNot one valuation. Every value the building could have, and how likely each one is.6 models
IIWhat the income will doWho pays the rent, for how long, and what happens to the loan when they stop.5 models
IIIHow to buy itWhat to ask, what to offer, who else is bidding, and what to do once it is yours.5 models
IVHow we knowWhere the evidence comes from, who attacks it, and how the Engine is held to account.7 models

§ IWhat it is worth

What it is worth

Not one valuation. Every value the building could have, and how likely each one is.

M-01Monte Carlo valuation

How likely is the asking price to be right, and how wrong could we be?

<1 in 100of ten thousand outcomes reach the asking price. Eight in ten land between £7.44m and £7.92m.

A single valuation is one guess. The Engine prices the building ten thousand times, each time drawing the rent, the yield, the empty months and the costs from what the evidence allows. Where the outcomes land is the value. How widely they spread tells you how much is really known.

Underneath

10,000 draws per building
Latin hypercube sampling
Each input drawn from its graded evidence: a weaker grade widens its draw
Rent and yield correlated, not independent
Re-run until the median moves by less than £5,000

Ten thousand values for the example building, with the asking price marked.Example Packet WC-0001 · hypothetical · illustrative

M-02Value sensitivity grid

How much does one small assumption move the price?

£330,000the value of a quarter of one per cent on the yield, on this building.

The yield is the return a buyer asks for, and small changes in it move the price a long way. The grid shows the value at every likely yield, on the rent paid today and on the market rent, so no single number can hide how sensitive it is. The asking price only works at 5.68%. Agents put buildings like this one at 6.0% to 6.5%.

Underneath

Net initial yield after 6.8% purchase costs
Passing rent and market rent, side by side
Reversion: the value when the rent moves to market
Every yield marked against the asking price

Value in £ million. Cells above the asking price are marked.Example Packet WC-0001 · hypothetical · illustrative

M-03Historical stress replay

What would 2008 do to this building? And 2020, and 2022?

Month 14when a repeat of 2008 would break the loan’s value test.

The Engine replays real crashes on the building and its loan, month by month. A repeat of 2008 breaks the loan’s value test in month fourteen. A repeat of the 2022 gilt shock breaks it in month five, and no recovery is assumed. A repeat of 2020 barely touches it: industrial values fell less than 3% that spring.

Underneath

2008 from all commercial property; 2020 and 2022 from industrial values, month by month
Rent held at the passing rent, the loan tested every month
No recovery assumed: a test that counts on one flatters the loan

Value of the example building through three replayed crashes, against the loan’s 70% value test.Example Packet WC-0001 · hypothetical · illustrative

M-04Supply and demand forecast

Will rents here rise or fall while you own it?

+2.4% a yearhow rents on buildings like this have moved here, measured building by building.

Rents are measured building by building across three rating valuations, so the change is what actually happened here, not a national average. Against it sits every new warehouse in the planning pipeline, set against the space that already stands. Here, the pipeline would add a fifth to local stock by 2028: the year before your tenant can leave.

Underneath

Rental change measured building by building, from the 2015, 2021 and 2024 valuations
New supply: every scheme in the planning pipeline, against local stock
Schemes counted by stage: consented, under construction, complete

New space in the local pipeline, set against the space that already stands.Example Packet WC-0001 · hypothetical · illustrative

M-05Climate and energy rules, to 4°C of warming

What will flood, heat and new energy rules mean for the building?

2031when a rented building of this size is proposed to need energy rating B, where the work pays back.

Flood, heat and rain are read for the building’s own patch of ground: the chance of surface flooding, days above 30°C and 35°C and warm nights at 2°C and 4°C of warming, and the driving rain each wall takes. Then the energy rules: a rented building over 1,000 m² is proposed to reach B in 2031. The cost of getting there comes from the certificate’s own recommendations or a quote, never a guess.

Underneath

Surface-water flood chance, in the official 1-in-30, 1-in-100 and 1-in-1,000 bands
Hot days, very hot days and warm nights, from national climate projections
Wind-driven rain on each face of the building
Energy rating against the proposed 2031 minimum

Days above 30°C on the example site: now, at 2°C and at 4°C of warming.Example Packet WC-0001 · hypothetical · illustrative

M-06Real options, priced by decision tree

What could the building become, and what is that right worth today?

£410,000of option value that the asking price leaves out.

The right to extend is worth something even if you never use it. It is priced as a tree of decisions: what it costs, what it could be worth, how likely each branch is. Here, spare land at the back allows a 20,000 sq ft extension, worth £410,000. A change of use is named, not priced: it turns on the power and the local plan. None of it is in the supported value. It comes free.

Underneath

Decision trees, valued backwards from each ending
Planning odds from every nearby industrial extension decision
Build cost from a quote; without one, the break-even

The extension on the example building, priced as a tree; a change of use, named.Example Packet WC-0001 · hypothetical · illustrative

§ IIWhat the income will do

What the income will do

Who pays the rent, for how long, and what happens to the loan when they stop.

M-07Default probability from filed accounts

How likely is the rent to stop, and what would it cost if it did?

7%chance the rent stops before the end of 2029. If it does, it costs £1.23m.

The tenant’s own accounts are run through two established failure scores, then set against how often firms in its industry really fail. The answer is not “strong” or “weak”. It is a probability, year by year, and it flows straight into the ten thousand outcomes.

Underneath

Altman Z″ score: 2.41, the grey zone
Taffler z-score: +1.4, solvent
Failure rates by industry and company size
Parent guarantee weighed on its own accounts
Cumulative probability to every lease event

The chance the tenant has failed, by each date, with its uncertainty.Example Packet WC-0001 · hypothetical · illustrative

M-08Clause extraction and dilapidations

What does the lease really say, and what is it worth?

£870,000of repairs the tenant will owe when the lease ends, before the legal cap.

Every clause that moves money is read and turned into a figure: the break and its conditions, the repairs the tenant owes, how the rent is reviewed. At the end of this lease, the repairs are costed at £1.04 million, and the 2019 record takes £170,000 of roof out of it. The law caps the claim at the fall in value; the Packet shows the claim, names the cap, and says what settles it.

Underneath

Break: only if the building is handed back empty, with all rent paid
Repair: full, limited by a 2019 schedule of condition
Rent review: upward only, open market, every five years
Repairs capped by statute at the loss in value

From the repair bill to the claim, before the legal cap.Example Packet WC-0001 · hypothetical · illustrative

M-09Whole-hold simulation, correlated shocks

Across the whole time you own it, how often do you make money?

1 in 4futures lose money at the asking price. At the opening price: 1 in 10.

Every quarter of a five-year hold is simulated: the tenant failing or leaving, the empty months, re-letting, the roof bill in 2029. Rent and yield are drawn together, not separately, because they move together. Nothing about the building changes between the two prices. Only what you paid.

Underneath

Quarterly steps, five years, ten thousand paths
Breaks and expiries at the rates industrial tenants have actually shown
Months empty stated as assumed, and tested at half and double
Rent and yield drawn together, not independently
Purchase and sale costs included

Money back for every £1 put in, across ten thousand five-year futures, at two prices.Example Packet WC-0001 · hypothetical · illustrative

M-10Cash flow with lender covenant tests

Does the rent still pay the loan in the year the roof bill lands?

0.94×how well the rent covers the interest in 2029. A lender wants 1.75×.

The loan is sized on today’s rent. Each big cost goes in the year it lands, and cover is tested in that year, not on average. Here, 2029 is the year the roof is due and the tenant can leave. Even if they stay, cover falls to 0.94×. If they leave, the rent stops. Pay for the roof at purchase and 2029 holds.

Underneath

Loan: 55% of value, 6.1% all-in, five years
Interest cover tested every year, never averaged
Capital costs placed in the year they fall
Two cases: the tenant stays, or leaves at the break

Rent against interest, year by year. Switch between the two cases.Example Packet WC-0001 · hypothetical · illustrative

M-11Deal clock, with collision detection

Which dates collide?

4collisions on this building that nobody had lined up.

Every date that matters goes on one line: the lease, the loan, the building, the market. Money is lost where they collide. Here, the rent review falls between exchange and completion. The tenant’s break sits inside the loan. A rival 120,000 sq ft warehouse opens a year before you might need a new tenant. And the loan ends six months before the lease does.

Underneath

Lease: reviews, breaks, notice dates, expiry
Loan: drawdown, tests, maturity
Building: repairs by the year they are due
Market: rival schemes by completion date
Power: connection dates offered

Five lanes, one line. Collisions are marked.Example Packet WC-0001 · hypothetical · illustrative

§ IIIHow to buy it

How to buy it

What to ask, what to offer, who else is bidding, and what to do once it is yours.

M-12Value of information

Which questions are worth paying to answer, and in what order?

126 to 1what a £3,000 check on the yield returns: it protects £380,000 of price.

Every open question has a price: how much the answer could move the value, times how likely it is to move it. Divide that by what it costs to ask, and the questions rank themselves. The building survey usually comes first. Here it comes fifth. A phone call comes first.

Underneath

Expected value of perfect and sample information
Every possible answer re-run through the ten thousand outcomes
Ranked by pounds protected per pound spent
Re-ranked every time an answer arrives

Six open questions on the example, ranked by what each answer is worth per pound.Example Packet WC-0001 · hypothetical · illustrative

M-13Seller reserve distribution and first offer

What is the least the seller will take, and where should you open?

£7.58mthe seller’s most likely lowest price. Eight in ten: between £7.38m and £7.79m.

The seller’s floor is worked out from what they owe, when their loan falls due, how long they have held, what their parent company needs, and the value in their own accounts. It is a curve, not a number. Open at £7,450,000 and stop at £7,850,000, and there is a 95% chance of a deal, at an expected £7.61 million.

Underneath

The seller’s debt, and when it falls due
Book value in their latest accounts
Fund life and holding period
The parent company’s cash position
Opening offer chosen to maximise expected profit, not to anchor

The likely spread of the seller’s lowest price, with the Packet’s three figures marked.Example Packet WC-0001 · hypothetical · illustrative

M-14Auction simulation

Who else will bid, how high can they go, and how often do you win?

64%your chance of winning at £7.85m. If the neighbour bids: 23%.

Each likely rival’s ceiling is worked out from what money costs them. Then the bidding is run ten thousand times. One buyer can pay more than anyone: the company next door, whose yard has no room to grow. If they know it is for sale, the asking price is right for them, and still wrong for you.

Underneath

Rival buyer types, from who bought buildings like this
Each rival’s ceiling from their cost of money
Special buyers found from the land around the site
Sealed bids and open bidding both simulated

Your chance of winning at each bid, with and without the neighbour.Example Packet WC-0001 · hypothetical · illustrative

M-15Exit buyer-pool model

When you come to sell, who will be there to buy?

2 of 5kinds of buyer would bid for it in 2031, as it stands.

A building is only worth what the next buyer will pay. The model asks which kinds of buyer would really bid when you sell, which would stay away, and why. As it stands, pension funds and listed companies stay away: the lease is too short and the energy rating too low. Fix both, and four of the five come back.

Underneath

Buyer types by lot size, age, lease length and energy rating
Each type’s yield on rateable value, from what it actually bought
The pool re-tested after each lever is pulled

Five kinds of buyer, as the building stands in 2031, and after the levers.Example Packet WC-0001 · hypothetical · illustrative

M-16Value levers, ranked

What should you do before exchange, and what after?

+£1.2madded by four moves after purchase, on top of £723,000 before.

Before exchange, the levers are the price and the contract: take the roof and the power off, or make the deal conditional on a connection date. After, they are the building: remove the 2029 break in return for six months free, sell the roof’s solar power to the tenant, join the queue for more power now, and insulate the roof in the same works as the coating.

Underneath

Every lever costed, timed and given an owner
Before exchange: price, conditions, warranties
After purchase: lease, energy, power, planning
Levers tested together, never simply added up

From the asking price, down to the supported value, and up again after purchase.Example Packet WC-0001 · hypothetical · illustrative

§ IVHow we know

How we know

Where the evidence comes from, who attacks it, and how the Engine is held to account.

M-17Director network analysis

Who is really behind the seller and the tenant?

1 directorsits on the boards of both the seller and the tenant.

The seller and the tenant are linked through their directors, past and present. That shows common control that ownership records hide, and “phoenix” companies that fail and start again under a new name. Here, the landlord and the tenant share a director, and the rent is above the market. The Packet flags it, and asks whether the rent was agreed at arm’s length.

Underneath

Directors and owners, current and resigned
Common control across groups
Phoenix patterns: failure, then a new name within months
Charges, guarantees and parent support

The seller and the tenant of the example building, joined by the people who run them.Example Packet WC-0001 · hypothetical · illustrative

M-18Sale comparables from public records

What have buildings like this actually sold for?

£66a square foot: the middle price of 62 sales of buildings this size nearby. The asking price is £140.

Every company purchase of a freehold nearby is matched to the building it bought, by street and number, never by postcode alone. That gives named, dated sales with a price per square foot. Here, 62 buildings this size have sold since 2021, the middle half for £35 to £95 a square foot. The asking price sits above almost all of them. The records do not say which sold let and which empty, which explains much of the spread, and the Packet says so beside the figure.

Underneath

Every company purchase of a freehold, matched by street and number, never by postcode alone
Leasehold sales listed and kept out of the figures
Land, shops and retail warehouses left out
Never a valuation: it shows where the asking price sits among real sales

Price per square foot of 62 sales of buildings this size nearby, with the asking price marked.Example Packet WC-0001 · hypothetical · illustrative

M-20Live re-pricing

What happens when one question is answered?

±£240,000the range today. Record an answer below and watch the whole file move.

A Packet is not finished once. Record one phone answer and the whole file updates: the ten thousand outcomes run again, the range tightens, and the questions left re-rank themselves. Try it.

Underneath

Answers weighted by their grade
Full re-simulation, never a patch
Open questions re-ranked by what they are now worth

Record an answer. The value, the range and the questions all move.Example Packet WC-0001 · hypothetical · illustrative

M-21Adversarial red team

Who tries to prove the Packet wrong?

142attacks on the example Packet. 131 failed, 9 moved a figure, 2 lowered a grade.

Before anything is signed, a separate stage, told only to prove the Packet wrong, attacks every figure and every grade. What survives is printed. So is what did not: the Document shows each attack and its result.

Underneath

A separate stage that sees what each room concluded, never how it was asked
Every attack must cite evidence
An attack can weaken a finding, never strengthen it
The full attack log is printed in the Packet

Some of the attacks on the example Packet, and what became of them.Example Packet WC-0001 · hypothetical · illustrative

M-22Forecast logging and calibration

How would you know if the Engine were wrong?

Everyforecast logged with its date and its probability, and scored when the truth arrives.

Every forecast in every Packet is logged: the rent review, the break, the price it sells for. When the truth arrives, it is scored. When the Engine says 70%, it should be right seven times in ten. Where it is not, its confidence is corrected. The record starts with the first Packet, and will be published once it is long enough to mean something.

Underneath

Brier and log scores on every probability
Calibration by tenth
Confidence recalibrated as results arrive
No record is shown until there is one

What calibration looks like. The record is empty until it is real.Example Packet WC-0001 · hypothetical · illustrative

M-23Monitoring after purchase

What happens once the building is yours?

Nightlythe building, its tenant and the land around it are checked again.

The Packet does not stop at completion. The tenant’s filings, insolvency notices, board changes and planning applications around the building are watched every night. When something moves, the figures it touches are run again, and you are told what changed and what it is worth.

Underneath

Filings, notices and board changes
Planning applications within 5 km
Every change priced where the evidence gives it a price

What a year of watching could look like on the example. Illustrative.Example Packet WC-0001 · hypothetical · illustrative

M-24Nightly off-market scoring

Which buildings will be for sale before anyone lists them?

Every nightevery industrial and logistics building in England and Wales is scored again.

The Engine does not wait to be asked. Every night it scores every industrial and logistics building in England and Wales on the signals that come before a sale: a loan falling due, how long the owner has held, a new lease, directors leaving, the owner’s own filings. When you ask about one building, it already knows it, and the ten like it that nobody has listed yet.

Underneath

Signals: debt falling due, how long the owner has held, a new lease, board changes, the owner’s filings
Scored nightly and ranked by the chance of a sale within 18 months
Matched to what you are looking for

An illustrative night. Each dot is a place; gold marks the likeliest to sell.Example Packet WC-0001 · hypothetical · illustrative

§ VAnd then

All of it, forone building.

Every model on this page is run on every building we take, and says plainly when the evidence cannot yet carry it. Their answers go into one Packet, every claim in it carries a grade, and one person reads every grade and signs.

These are some of the models, not all of them. You do not need to follow the mathematics to use the answer. That is the point of it.

Submit an asset