No Reference Point: Deciding Without a Distribution | HL Hunt
No Reference Point: Deciding Without a Distribution
To know whether a price is good you need the distribution it came from, and households almost never have one. They see one number at a time, and a single observation carries no information about where it sits. Six quotes for identical work spanned 64% of the median; one quote has a 17% chance of being the lowest. The useful finding is where the benefit concentrates: the second quote is worth $530 and the third only $208.
What you'll learn
The core thesis
To know whether a price is good, you need the distribution it came from. To know whether a rate is high, you need to know what people like you are offered. To know whether a fee is normal, you need to know what normal is.
Households and small firms almost never have any of that. They see one number at a time, and a single observation carries no information about where it sits.
Our position is that the absence of a reference point is a distinct financial disadvantage, separate from income, credit quality, or knowledge, and that a great deal of what looks like poor decision-making is a reasonable response to having one data point.
Per our constraint analysis, this is not a knowledge deficit that education fixes. Knowing that you should compare prices does not supply a distribution, and the cost of generating one yourself is the thing being economized on.
What one quote is worth
Take six quotes for the same piece of work: $2,850, $3,100, $3,400, $3,950, $4,300, $5,200.
- Median: $3,675
- Spread: $2,350, which is 64% of the median
- Chance a single quote is the lowest: 17%
Now the value of each additional observation:
| Quotes obtained | Expected best price | Gained |
|---|---|---|
| One | $3,800 | |
| Two | $3,270 | $530 |
| Three | $3,062 | $208 |
That concentration is the practically useful finding. The advice to shop around implies an open-ended search. The arithmetic says the decisive move is the second data point, after which you are refining rather than discovering, and per our exception analysis the effort required for each further observation is the same while the return falls.
It also explains a behaviour that looks irrational and is not. A household taking the first quote is forgoing about $950 relative to the best available, which is a real loss and a defensible one if obtaining the second quote costs time it does not have. Per our latency analysis, the cost of search falls hardest on the people whose hours are least their own.
Where the reference point is missing
| Decision | What a reference point would tell you |
|---|---|
| A loan rate offered | Whether it reflects your file or the lender's margin, per our intermediary analysis |
| A card processing rate | Whether 2.9% is the market or the markup, per our pricing analysis |
| An insurance premium | Whether the increase is the market or your renewal |
| A professional fee | Almost nothing is published, so the first number is the only number |
| A supplier price | Whether you are being charged a small-account premium, per our supplier terms analysis |
| Your own margin | Whether 38% is good in your trade, per our quoting analysis |
The last row is the one that compounds. A business that does not know its trade's typical margin cannot tell whether its pricing problem is a pricing problem, and per our capacity analysis every decision built on that unknown inherits the error. The firm is not short of data about itself. It is short of anything to compare that data to.
The third row illustrates the asymmetry cleanly. The insurer knows the distribution of premiums for customers like you. You know your premium last year. Both parties are looking at the same decision with radically different information, and nothing in the process corrects that.
The substitutes people use, and what they cost
- Last year's number. The most common reference point and a poor one, because it anchors on your own history rather than on the market, so a price that drifted up for years looks normal.
- What a friend paid. One other observation, of uncertain comparability, which is nonetheless the single highest-value move available per the arithmetic above.
- An advertised rate. Per our intermediary analysis, advertised rates are the best available to the best applicants and tell you little about your own offer.
- An online average. Better than nothing, usually not conditioned on anything about you, and frequently published by someone with an interest.
- The provider's own claim that the offer is competitive, which is the reference point most often accepted and the least independent.
Item one is worth singling out because it is invisible as an error. Anchoring on your own prior price makes any gradual increase acceptable and any gradual market decrease invisible, and per our pricing analysis suppliers understand this well enough that incremental annual increases are the standard approach to repricing.
A falsifiable claim
Variation in prices paid for identical services should be substantially larger among customers with fewer observations, and the effect should be separable from credit quality and income. Anyone holding transaction-level pricing data can test it: for a standardized product, compare the spread of prices paid against a proxy for search, such as the number of applications or quotes recorded. If the spread is similar regardless of search intensity, prices are determined by customer characteristics and the reference-point argument adds nothing. We expect a clear relationship, which would mean a measurable share of price dispersion reflects information rather than risk, and is therefore recoverable by publishing distributions rather than by changing anything about the customers.
Why distributions are not published
The absence is not accidental and the reasons are not all self-serving.
- Prices are genuinely conditional. A rate depends on the file, the term, the security, and the channel, so a bare average can mislead more than it informs.
- Publishing invites convergence. Per our monoculture analysis, a published benchmark becomes a target and prices cluster at it, which removes the low end as well as the high.
- Competition concerns cut both ways. Price transparency between competitors is in some contexts a problem rather than a remedy.
- Nobody holds the data. The distribution of what households pay for a service exists only in aggregate across firms that have no mechanism or reason to pool it.
- Averages hide the relevant question. The useful figure is the distribution conditional on being like you, which is far harder to produce than a mean.
The second bullet is the strongest objection and should temper any enthusiasm. A published benchmark can raise prices at the bottom as much as it lowers them at the top, because firms pricing below it discover they are leaving money on the table. Transparency is not unambiguously good for buyers, and anyone arguing for it should be able to say which direction they expect the effect to run.
What to do with one data point
- Get the second one. It is worth $530 in the worked example and the third is worth $208, so the single highest-return action is going from one to two.
- Ask the provider what range exists and on what basis it varies. A provider that will not describe the range has told you something.
- Separate the components. Per our pricing analysis, a bundled number cannot be compared and an itemized one can.
- Anchor on a calculation rather than on last year. What the thing should cost from its inputs beats what it cost you before.
- For a business, find the trade figure. Industry associations publish margin and cost benchmarks that most members never look up.
- Record what you paid and what you were quoted. Per our institutional memory analysis, your own history becomes a distribution after a few cycles, which is the only reference point you can build without anyone's cooperation.
Item six is the slow answer and the only self-sufficient one. A household or firm that records quotes it did not accept is building the distribution it currently lacks, and after three or four renewal cycles has something no provider supplied and nobody can withdraw.
Scenarios and what we are watching
| Scenario | What we would expect |
|---|---|
| Comparison tools improve | Dispersion falls, and per our intermediary analysis the ranking incentive determines whether that helps buyers |
| Conditional distributions get published | The genuinely useful version, and the hardest to produce, because it requires pooled data nobody holds |
| A benchmark becomes a target | Clustering at the benchmark, which compresses dispersion from both ends |
| Nothing changes | A measurable share of price dispersion continues to reflect how many quotes someone had time to get |
The last row is the claim this article rests on and the one worth stating plainly. If prices paid for identical services vary by 64% and a meaningful part of that variation tracks how many observations the buyer had rather than anything about the buyer, then some of what is recorded as different customers getting different prices is the same customer getting a different price depending on how much time they had that week. That is not a market outcome anyone designed, and per our legibility analysis it is the kind of cost that never appears in any ledger because no line item corresponds to it.
Frequently asked questions
In the worked case, six quotes for identical work spanned $2,350, or 64% of the median, and one quote has a 17% chance of being the lowest.
The second quote is worth $530 and the third $208. Almost the entire benefit is in going from one observation to two.
No. Knowing to compare does not supply a distribution, and generating one costs time. Anchoring on last year’s price makes any gradual increase look normal.
Prices are genuinely conditional, nobody holds the pooled data, and a published benchmark becomes a target that can raise low prices as much as it lowers high ones.
Key takeaways
- A single price observation carries no information about where it sits in a distribution.
- Six quotes for identical work spanned $2,350, or 64% of the median, with a 17% chance any one of them is the lowest.
- The second quote is worth $530 and the third $208, so the decisive move is going from one observation to two.
- Taking the first quote forgoes about $950, which is a real loss and a defensible choice if the second quote costs time you do not have.
- Anchoring on last year’s price makes gradual increases look normal and gradual market decreases invisible.
- A business that does not know its trade’s typical margin cannot tell whether it has a pricing problem.
- The strongest objection to publishing benchmarks is that they become targets, compressing dispersion from both ends.
- Recording quotes you did not accept builds, over a few cycles, the only distribution nobody can withdraw.
This analysis is educational and does not constitute financial advice. All figures are stylized illustrations using an assumed set of quotes, and expected-value calculations assume each quote is an independent draw from that set, which is a simplification. Price dispersion, the availability of published benchmarks, and the comparability of quotes vary substantially by market and by product.