AEM 6991

MPS Capstone Project I

Prof. Ariel Ortiz-Bobea

3 · Finding the question

Wednesday, September 9, 2026

1. What makes a good idea

What makes a good idea

• You need a way to quickly evaluate ideas, otherwise you hang on to less interesting ideas for too long.
• An good idea has to check a few boxes:

1. Importance. Somebody is better off knowing the answer.
2. Feasible. One can actually get the answer with existing data and methods.
3. Novelty. The answer is not already known.
4. Fit. The idea is great for your team. You have the skills and comparative advantage to tackle it.

• You can rank ideas based on these criteria to pick ideas with high potential for your team.

One and a half minutes. Put the four up and do not explain them yet. One slide each follows.

Say why this comes first. You already have ideas, so the useful thing is not another way to generate them. It is a way to sort them fast. The cost of having no test is not that you pick a bad idea on purpose. It is that you stay with a mediocre one for weeks because nothing told you to drop it.

Say that the four are not weighted equally in this course. Feasible is the one that kills projects in March, so it gets the whole second half of the hour.

Say also that they pull against each other. The most important questions are usually the least feasible, and the most feasible are usually the least new. A project is a choice about where to sit between them, not a search for a question that scores full marks on all four.

The fourth is the one they will not expect and it is the one that stops a team adopting somebody else’s good idea. Flag it here and let the fourth slide do the work.

What makes a question important

• Answering the question provide value to someone.
• The value you bring can be a private or a public good.
• Private good: e.g. you shed light on an untapped market segment for a company (value is derived by acting upon the private information, not publishing it)
• Public good: e.g. you shed light on how changing a policy could improve economic surplus (value is derived by publishing)
• In academia, the important is often measure by how many other academics find the work useful (determines journals and drives citations)
• Key idea: someone is better off after we responded to this question.

Ninety seconds. This is the test that the last two cohorts failed, and it is the one the room will resist, because a big topic feels important.

Push on the difference. Climate change is large. Whether a flood changes what a lender charges is important, because somebody acts on the answer.

Do not resolve who acts and who pays here. That is the next block, and it is where this test gets its second half.

What makes an answer valuable

• Somebody or some organization has a decision to make that would benefit from it, or a whole agenda cannot move without it.
• If you cannot identify the people or organizations benefiting from the new knowledge, then it’s unclear this is important.
• Public good value is shared: many people are better off and you cannot stop anyone using the answer, so it is hard to charge for and it gets published.
• Private good value can be kept: one person’s decision changes, and that person can keep the answer to themselves, so somebody pays for it.
• Both are fine for this project. Private value has greater traction in private industry.

Two and a half minutes. This is the fourth filter, would anybody act differently, arriving early. Say the word value and leave the slide up.

The bullet that says the test is whether the answer can be kept is what makes quiz item 1 work, because a student who learned “private means somebody acts” picks the wrong distractor without it.

Neither definition uses the word excludable. Say “you cannot stop anyone using it” out loud too, because the term is not what is being examined.

The theory behind the public half is Arrow 1962 on knowledge as a non-excludable good, cited on the companion page and not on the slide.

Say plainly that the grade does not prefer private.

Who acts on the answer or pays for it

1 · attention and stock prices · private value
An investment fund would pay for this while it is the only one holding the answer. Ask instead whether small investors lose money and your reader is a regulator, which is a different project.

2 · extreme weather and damage · both
Finance ministries and the World Bank act on a published number. Insurers would pay for the same number, because it is what they price the cover on.

One minute. The lesson is the sentence under each label, the one that names who acts or pays. A team that can fill it has passed filter 4. Read both of those sentences out; the labels can be said in one word each.

Each block names its item in three or four words, because the vote slide is gone by now and “1 · private” on its own tells a student nothing.

Insurers on the screen is reinsurers in fact. Say reinsurers if you want, after the plain word, and do not spend a sentence on the difference.

The plan draws this reveal as a three-column table. It is set as two stacked rows instead, because the first “why” cell is twenty-four words and a cell that long does not fit at this root size.

Item 1 is Team D’s sentence shown as a buyer it has, not as a verdict. The verdict on signal-on-returns sits on “Private value is easier to sell, and it evaporates when published”, one slide from here.

Gate. This slide is on screen by 2:24. If it is not, read it as one item only, and drop “Private value is easier to sell, and it evaporates when published” to its first and last bullet. That saves a minute and a half.

Two minutes. Start the count-down at thirty, projected beside the slide, and say nothing until it ends. It shows this write’s own seconds and it is the only thing that ends the write. Write, do not talk. Nobody reads one out.

This line is the pitch’s why-it-matters line, written while the examples are on the screen. It goes on the question sheet, the slate on the printing list, this afternoon in the same words, which is why the last bullet exists: the strip is a personal half sheet that is never collected, and it is the thing most likely to be left on a chair at 3:15. The same instruction is printed at the top of the slate sheet, which the sheet itself calls the question sheet.

Transition line: value says why, and the shape says what you can claim.

What makes a question feasible

• You can get the data, in the time you have, with the skills on the team.
• Feasibility is about resources. It is not about whether the question could be answered by somebody.
• Most projects fail here, so it is critical to evaluate this as early as possible.
• A check: can you pin point the data you are using (if using secondary data) or have an idea of the platform to run your experiments (if working with primary data).

Ninety seconds. Say plainly that this is the one that does the killing, and that it is why the second half of the hour is one long exercise in it.

Checked, not assumed, is the phrase to repeat. It comes back on the map as the difference between a have and a need.

The whole resource map is this test taken seriously, so this slide is its promise. Do not open the map here.

What makes a question novel

• Nobody has already answered it, and you know that because you went and looked.
• Requires that you read and look around extensively.
• New does not mean nobody has worked on the topic (that could be a sign this is not important). Instead, it means this answer is not already out there (e.g. a publication, a product or service line, etc.).
• A marginal change to published work is new, but not as important (you need to recognize this).
• The check: what is the closest study to ours, and clearly identify how your study sheds light on something the previous study did not provide any insight.

Ninety seconds. This is the test that sends them to the literature, and it is the only reason this course asks them to read.

Tie it back to the second route without naming a team. A marginal change passes this test at the low end, and that is acceptable for one semester as long as the team knows where it is sitting.

The honest failure here is not being unoriginal. It is not having looked.

Private value disappears once it is published

• McLean and Pontiff collected 97 published rules for picking stocks that beat the market.
• On average a rule earned 26 percent less on later data than in the study that found it, and 58 percent less once it was in print.
• A rule is worth money while one trading desk is the only one using it, and about half that once it is in a journal.
• A search-traffic signal regressed on next-day returns has no trading desk and no journal, so it has neither kind of value.
• If an answer is worth money and nobody has produced it, ask why, and read to check that nobody has.
• A public good does not lose its value like this.

McLean and Pontiff (2016), Does Academic Research Destroy Stock Return Predictability?, Journal of Finance 71(1), 5 to 32

Ninety seconds. This is why signal-on-returns is a fine desk memo and a bad capstone.

Nobody in the room has to have read a finance paper. Return predictor is “a rule for picking stocks that beat the market”, out of sample is “on later data”, and both are said that way on the slide.

The figures and the DOI are verified, so they go on the slide as numbers and not as “about”. Do not say “near zero”: it was wrong and contradicted the paper cited one bullet above it.

The last bullet is meeting 1’s line on what reading is for, turned into a reason to go to the literature.

What makes a question a good fit for you

• A question can be important, feasible and new, and still not be a good fit.
• Fit is two things: the skills your team actually has, and whether you care about it.
• A question nobody on the team can execute is somebody else’s project, however good it is.
• The check: do we know the skills we have in the team? Do we have the excitement and commitment for this? Are we well organized and know who does what?

Ninety seconds. This is the instructor’s own fourth test and it is the one that is missing from every version of this advice they will read elsewhere.

Say the failure case plainly. A team adopts a question because it sounds important, nobody on it can do the hardest part, and by February the project is being done badly by the one person who can.

Interest is not softness. The question that gets finished is the one somebody still wants to answer when it has stopped being fun, and that is a feasibility fact rather than a preference.

This is also the argument for the inventory on “Start from what you have”, forty minutes from now. Do not open it here, just plant it.

2. Types of ideas

A project can take different shapes

	The question it asks	For example
1 Evaluation	Did it work? Somebody changed something and you measure what it did	Did the loyalty card raise what people spend?
2 Preference and pricing	What would people choose and pay? Nobody has decided yet, so you ask them	Would shoppers pay more for a lower carbon label?
3 Prediction and risk	What happens next? You need the number, not the reason	Which customers will cancel next quarter?
4 Diagnosis	Why is this so uneven? Something varies and nobody knows why	Why do some stores sell twice what similar stores do?
5 Feasibility	Is this opportunity real, and what is in the way?	Is there a market for this product in Kenya?
6 Measurement	How much of this is there? Nobody has counted it, so you build the number	What does one flood actually cost a county?
7 Valuation	What is this worth? You price it against things that already carry prices	What is a water view worth on a house?

• The same interest can usually be turned into three of these, so pick the one you can answer.

Three and a half minutes. This slide defines the five. The next one kills them, so do not give away the failures here.

Read the middle column out. It is written for somebody who has never taken an econometrics course, and the examples are there so nobody has to guess what the label means.

Rows six and seven were added on 2026-09-09 because five did not cover the room. Measurement is how much of this is there when nobody has counted it, and it is where the climate damage question sits. Valuation is what is this worth, priced against comparable things, and it is where the asset pricing question sits. Neither team is named on the screen.

The old version of this slide showed only the failure column, which named five categories it never defined. It was cryptic and it was cut on 2026-09-09.

The can-claim column is spoken rather than shown, and all four columns are printed on the companion page. Say the one line that matters: one and two can support a claim about cause, three cannot and must not pretend to, four is a description that points at a cause somebody could test later, and five needs at least one piece of the team’s own evidence or it is a literature review.

Then say Team H’s sentence as three shapes, as a gift. The sentence is “does AI advice move a non-expert closer to a good decision, or just swap their overconfidence for the model’s?”. Evaluation: did people given AI advice decide better than people without it? Preference: which advice format do people choose, and what would they pay for it? Prediction: who over-relies, and can you tell in advance?

A team that hears its own interest in one of those three has found its afternoon’s first job.

3. How you come up with ideas

Ideas arrive in three ways

1. From the world. Something you noticed, or a decision somebody has to make.
2. From previous work. A marginal change to something already published.
3. From data. A file you can already reach, worked backwards to a question.

• One slide each follows, and each one ends with the thing that route leaves missing.

One minute. Put the three up and take them one at a time afterwards.

Say the shape of the block before it starts: the world lacks data, previous work lacks novelty, data lacks a decision. They will hear each of those again on its own slide, so do not explain them here.

Meeting 1’s page calls the data route “route 5”. Say once, and only once, that the five collapsed into three, because a gap and an extension are both previous work, and an anomaly and a decision are both the world.

Getting ideas from the world

• From the world: a problem you saw, and it usually arrives without being sought.
• It comes from the news, from a conversation with a colleague, or from a job you have done yourself.
• The good ones often come out of an unexpected conversation rather than out of sitting down to think.
• You start with the problem and go looking for data afterwards, so data is the row most likely to be empty.

Ninety seconds. This is the route most of the room arrived on.

No team is named on this slide or on the two that follow. Naming one makes the route a verdict on their question, and the three routes are not a ranking.

Say once, and only once, that meeting 1’s page calls the data route “route 5”, and that the five collapsed into three, because a gap and an extension are both previous work, and an anomaly and a decision are both the world.

Gate. This slide is on screen by 2:11. If it is not, run all three route slides at a minute each by dropping the third bullet from every one.

Getting ideas from previous work

• From previous work: you take an existing idea, generally published, and change it at the margin.
• The change is usually a new setting, a new period, or a limitation the original authors name themselves.
• Read the limitations section, because that is where the authors write down what they could not do.
• This route tends to be marginal, and saying so is honest rather than modest. For one semester it is the right size.

Ninety seconds. Marginal is the instructor’s own word for this route and it is not a criticism. Say it plainly.

The data usually exists already, because the first study needed it. What is missing here is novelty, and novelty is not a row on the map, which is why the reading has to happen before the data work rather than after it.

Do not say which team in this room is on this route. Several are.

Getting ideas from data

• From data: you start from a file you can reach and work backwards to a question it could answer.
• Data is usually the binding constraint, so this route makes the constraint visible on day one.
• Expect to go back and forth. You get the file, find what it does not contain, and revise the question.
• That revising is not the plan failing. It is the loop the rest of this hour is about.

Ninety seconds. This is the instructor’s own account of the route: the data constrains the question, and the question moves.

Gloss Dewey here, in one clause, because it is named on the access ladder forty minutes from now and nowhere in between: the data marketplace the business school subscribes to, free to them, and approval takes days.

The reverse-engineering is the next slide, so do not deliver the warning here. The fourth bullet is its set-up, and it is also the first statement of the loop the resource map draws.

Working backwards from data honestly

• What could this data answer?
• Which of those would anybody act on?
• What does it not contain that you would need?
• You are allowed to walk away from the data.
• Starting from data is the only route where the data is already in hand, so nothing looks missing, and only the question of who would act can kill it.

Ninety seconds. The first four lines are meeting 1’s, said once and not re-taught.

The fifth is the new thing and it is why the slide survives. It is the hook into “What your question needs against what you have”, and it is the tell the shared case turns on after the break.

The map words belong to that slide and not to this one, so do not say “the data row is full and the gap column is empty” here. Say what the bullet says.

If the 2:16 gate is missed, the whole slide becomes one spoken sentence on “Which route did your team’s question come from?” and the fifth line is the sentence.