Part IIB, Paper 4, Economic Theory and Analysis

Lent 2024

Aytek Erdil

Lecture 3

6 February 2024

1 / 14
Many-to one matching

Extend the theory to account for colleges with multiple seats.

S: a nite set of students

C: a nite set of colleges College c has qc seats

Student s has a ranking %s over C ∪ {s}

College c has a ranking %c over S ∪ {c}

Matching µ is a correspondence from S ∪C to S ∪C such that

I µ(s) ∈ C ∪ {s},

I µ(c) ⊆ S such that |µ(c)| ≤ qc ,

I µ(s) = c ⇐⇒ s ∈ µ(c), for every s∈S and c ∈ C.

2 / 14
Colleges' preferences over groups of students
Note that for the colleges, the model's primitives specify preferences over
individual students only (a rank order over S ).
In reality, the colleges must be comparing sets of students, perhaps based
on a preference ranking over all subsets of S.
S
Let's denote by 2 the set of all subsets of S.
S
Of all possible preferences over 2 , we restrict attention to those which
are responsive dened as:

I whenever T ⊆ S\{s, s 0 }

T ∪ {s} %c T ∪ {s 0 } ⇐⇒ s %c s 0

I whenever T ⊆ S\{s}

T %c T ∪ {s} ⇐⇒ c %c s

and
T ∪ {s} %c T ⇐⇒ s %c c
3 / 14
Group stability and pairwise stability

A matching µ is blocked by a coalition B of colleges and students if

there exists another matching µ0 such that

1. µ0 (s) ∈ B for all students s∈B

0
2. µ (s) s µ(s) for all students s∈B
3. s 0 ∈ µ0 (c) ⇒ s 0 ∈ B ∪ µ(c) for all colleges c ∈B
0
4. µ (c) c µ(c) for all colleges c ∈B

µ is group stable if it is not blocked by a coalition of any size.

A weaker stability concept is pairwise stability which requires robustness

to blocking coalitions of size 1 or 2 only.

4 / 14
Pairwise stability
A blocking coalition of size 1 means

I a student who is matched with a college she nds unacceptable, or

I a college which is assigned a student it nds unacceptable.

A blocking coalition of size 2 means a student-college pair (s, c) who are

not matched, but would rather be matched via the college oering an
empty seat (if exists) to s, or replacing a lower ranked student with s.

In other words, µ is pairwise stable if

1. µ(s) %s s for all s ∈ S; and s %c c for all s ∈ µ(c),

2. There are no s∈S and c ∈C such that s c c , µ(c) < qc , and
c s µ(s),
3. There are no s, s 0 ∈ S and µ(s 0 ) = c ∈ C such that c s µ(s) and
0
s c s .

5 / 14
Theorem. A matching is pairwise stable if and only if it is group

stable.

Remember:

Robustness to individual blocks 

(i.e., individual rationality)



pairwise stability
Robustness to pairwise blocks




(i.e., no blocking pairs condition)


Robustness to blocks of any size } group stability

Exercise. Prove the theorem. (You should be relying on responsiveness of

colleges' preferences to show that pairwise stability implies group

stability.)

6 / 14
Stable matchings always exist in many-to-one matching
Theorem (Gale and Shapley, 1962)
In every many-to-one matching market, a stable matching exists.

Step 1: Every s applies to her favorite acceptable college. Each college c

places the highest ranked qc applicants among the acceptable ones (all
applicants if there are fewer than qc ) on its hold list, and rejects others.

Step k: Applicants who were rejected at step k −1 apply to their next

best acceptable college. Each c places the highest ranked qc acceptable
students among the new applicants and those in the hold list on the new
hold list, rejects the rest.

The algorithm ends when every student is either on a hold list or has
been rejected by every college that is acceptable to her.

At the end, colleges are matched to students on their hold lists.

Exercise: Write down the college-proposing version and prove that the

outcome is stable.

7 / 14
One-to-one vs. many-to-one environment
Think of each college c as qc dierent colleges with one seat each.

Assume preferences are strict.

Many, but not all, results continue to hold.

1. Student-proposing DA results in the student-optimal stable

matching.

2. College-proposing DA results in the college-optimal stable matching.

3. For every college c, the number of students matched with c is

constant across all stable matchings. The set of matched students
is the same across all stable matchings. (In fact, an even stronger
statement holds. See the supervision problem set for that
statement.)

4. Student-proposing DA is strategy-proof for students.

5. However, no stable mechanism is strategy-proof for colleges.

8 / 14
An application: school admissions
Each student i ∈N has strict preferences over schools and her outside
option.

For each school x ∈X there is a priority ranking %x over students.

(Perhaps based on a catchment area, siblings, etc.)

All students are acceptable to all schools.

A matching violates student i 's priority for school x if i would rather

be at school x than her current match, and

I school x an empty seat, or

I another student j is matched with school x even though j has less

priority for x then i

A matching is individually rational if every student prefers her matched

school to her outside option.

9 / 14
An interpretation of stability in this context: respecting
priorities
A matching µ respects priorities if

I it is individually rational (IR), i.e.,

µ(i) %i i for each i ∈N

I it does not violate priorities, i.e., there is no pair (i, x) such that

I x i µ(i) and µ(x) < qx , or

I x i µ(i) and j ∈ µ(x) and i x j

Violating priorities is mathematically equivalent to the existence of a

blocking pair.

So respecting priorities is technically equivalent to stability if priority

rankings are treated like college preferences.

10 / 14
Interpretation, however, is dierent
Priorities are determined by local laws.

They do not necessarily reect the schools' preferences, and are not
chosen strategically (or otherwise) by the schools.

Therefore

I only student preferences constitute the welfare criteria, and

I students are the only strategic actors

Question 1: Is there a matching which

I respects priorities,

I is optimal from students' perspective?

Question 2: If so, can we implement (bring about) that matching?

11 / 14
With strict priorities

If priority orders of the schools are all strict, then we know that the
student-proposing DA

I yields the unique student-optimal stable matching

That is, the matching which respects priorities, and Pareto dominates all
other stable matchings from students' perspective.

Moreover, we know that if priorities are strict, the student-proposing DA

is

I strategy-proof for students

Hence, we can implement (in dominant strategies) the desired outcome

with a direct revelation mechanism.

12 / 14
In practice, however ...

In reality, school priorities are often coarse.

For example, in England whether you live in the school's catchment area
or not determines your priority as high versus low.

And perhaps having an older sibling at the school gives the student
higher priority.

Those two criteria would lead to a ranking with only four priority classes.

Large groups of students will be in a tie.

We can still use the student-proposing DA to nd a matching which

respects priorities:

Arbitrarily break ties in priorities and carry on.

13 / 14
Next time

We will explore how ties matter in this problem and whether we can
address the related issues.

We will look into the tension/conict between student welfare and

respecting priorities (where student welfare means Pareto eciency from
the students' perspective).

14 / 14