Welcome to the transcript of OddChain’s conversation with top Kalshi Trader Gaëtan Dugas. Here we discuss:

Let’s dive on in!

This transcript has been edited for clarity and length.

Welcome to the inaugural episode of what will eventually become the OddChain podcast. For now, we're going to be interviewing some of the top minds and traders in the prediction market space. To that end, we're very excited to welcome as our first guest, none other than Gaëtan Dugas. I really appreciate having you here, Gaëtan.

Hey, I didn't realize I was the first guest ever. What a tremendous honor.

Introduce yourself to the listeners a bit. Want to tell everybody who you are? How long you've been around in prediction markets?

My avatar, my nickname I go by online is Gaëtan Dugas. I guess, I've been doing it for a little over ten years now. I started on PredictIt back in 2015. I mostly did politics up until like 2021 or so when Polymarket started to offer more options.

I'm probably about a Top 100 trader on Kalshi right now. I've made $581,000 on Kalshi, and I've only been doing it for a little under two years on Kalshi.

And then, regarding my background, I studied finance, and I got an MBA as well. And then I keep my day job as a business analyst. It's an IT role. I'm basically telling developers what to do.

I love that. Do any of your coworkers know what the night shift gig is or do you keep that to yourself?

Yeah. Like, when I bring it up, or if we're in a meeting and I bring it up, it becomes the topic of the rest of the meeting. They're pretty fascinated by it. And then I'll send them The Washington Post articles and The New York Times and stuff like that as well.

That's so cool. It certainly makes for good water cooler stuff, I'm sure. But one of the things that I appreciate about your presence is that I think you're one of the social media figures who's really good at giving good advice to newcomers. 

You talk a lot about how to build models for markets, and about really data-backed theses for these things and how to find edge with those models. Can you go into a little bit more detail about what that actually means from a practical perspective?

Sure.

Like, if you have an edge, how do you build to get that?

Sure. So, I'll walk through Spotify as an example since it's pretty easy compared to a lot of the other ones and the data is so easily accessible.

So with Spotify, they've got their whole daily history of streams of the Top 200 for global and in every country, going back to like 2017. So, you've got access to this entire database of streams. And you can pull the data and start analyzing it, and you'll start to recognize patterns.

So, there are patterns that happen when a song is released. Friday is typically the biggest day. And then Saturday, it will drop by a certain percentage. Sunday it will drop another percentage. On Monday it usually bounces back. On Tuesday it bounces back a little more. And then from there, it does a slow decay. So you can spot that type of data. And you can look for comparable songs in the history to new ones that are coming out and kind of map out what and how you think it will do.

So, when it comes to modeling, I've got different models for different types of songs. A song like "Please, Please, Please" by Sabrina Carpenter or "we can't be friends" by Ariana Grande. Those are songs that did very well; they didn't drop very much.

And then there are songs that kind of flop, like "Aperture" by Harry Styles was a recent one where it dropped a ton.

And then there's kind of the average one.

So, basically what I do is I try and come up with a range of how songs will perform, and then once I track it a couple days, then I can get a pretty good idea of which of the categories it falls into. And based on that data, I can start projecting how it's gonna do the rest of the week.

Does that make sense as an example?

Yes, I can kind of see the spreadsheet for that too, right? You sort of slot it in. When you talk about building a model, sometimes people build more of a mental model, but it seems like for you it's very numbers-based. I imagine you sort of run these songs through the Excel sheet?

Yeah, so I'll run the numbers in Excel and get a high, medium, and low range for what the streams will be throughout the week, and then I just narrow that down as more data comes in.

So, that's a fairly simple one, but for something like Rotten Tomatoes, that's actually about six models that run at different points throughout the week because the amount of data that you get is different. So, before the social media embargo lift, all you can really find is random people rating it on Letterboxd, IMDb, or sometimes Google, or pulling reactions from Twitter. But as you get closer to release, you get more and more data. And then once reviews come in, you need to account for all the reviews that you can find on websites that haven't been published yet and how likely they are to be posted to Rotten Tomatoes in time. So, I take all that information and each step of the way has a different model that applies to it.

That's very cool. I've seen you talk too, though, that sometimes you do bets for fun, kind of based on gut feel. I'm wondering if the two betting styles ever overlap. Like, you have the model, but every now and again you're like, "No, this song's a banger. I bet it actually does better than the model says."

Yeah. So, today's Thursday and at midnight Eastern they're going to drop some new songs, like the J. Cole album's gonna drop. And if something big is coming up I'll usually stay up until midnight and just listen to them. And at that point, you don't really have any data on that specific song, so it really is a matter of listening to it and seeing if it's going to be any good or perform well. But for something like J. Cole, where it's an album people have been waiting for, I'll go back and compare how he did on his last studio album, which I think was in 2021. He did a mixtape release in 2024, as well. So I'll see how he did in the past and then consider if his popularity has dropped or if it's risen. I think it's probably dropped a little, but not enough that he won't get number one for the day tomorrow. So, it's a little bit of gut and a little bit of research on how that particular artist has done.

I like that a lot. A mixture of hard data and cultural analysis.

Yeah, for sure.

But yeah, I think that's a good overview of what it means to build a model. Let's say somebody's just starting out today. They've probably lost some money gambling on sports or whatever and they're trying to get more sophisticated. Where do you think the highest upside is these days? Where are the most inefficient markets? Where could somebody actually go out and find edge, in your mind?