How to Measure Liquidity in Prediction Markets

I’m currently 4 days into learning about day trading and I’m struggling so hard on understanding what liquidity is.

I’m watching TJR, & other people. But once I got to Liquidity i completely folded and don’t understand a single thing whatsoever and it’s hard for me to comprehend what ANY of the people trying to teach me is saying for some reason.

Liquidity is how easy it is to buy or sell an instrument. "An illiquid instrument will be hard to buy or sell" "A liquid instrument will be easy to buy or sell"

Liquidity is not -

• a place on a chart
• an area where you think stops are
• places where you think traders might enter or exit
• it does not exist in the past, it is only current and relative to last traded price

In case you didnt know, TJR is a child with no market experience or knowledge. He is a clickbaiter and he makes money from clicks and views like any other social media influencer.

With Google dropping 85 billion of new shares, SpaceX getting ready to IPO and sell 75 billion, and Anthropic and OpenAI filing their S1s to go public "soon" there's at least 160 billion that needs to be "bought up" if the IPOs sell at their target price, and a total of 320 billion if we assume Anthropic and OpenAI choose to raise similar amounts (85 + 75 + 80 + 80 billion)

The largest year of IPOs so far has been in 2021 with $303 Billion done in a year. This means this year will likely top that by around 20 billion

We did it in 2021, albeit when interest rates were low and money was cheap, but is this a significant amount of capital for the total public market? Or is it large but not much more than a drop in the bucket in terms of total public market?

Obviously the money has to come from somewhere, but understanding the size of the ocean helps measure the size of the drought

I know this might be basic, but I’m still trying to understand how mm's trade the other side of say equities during strongly directional events (e.g., bullish earnings or major news).

For example, during a clearly bullish earnings release - where there’s overwhelming demand to buy something like Apple, a mm could end up accumulating a large short position as they fill buy orders. In that scenario, they’re effectively taking the other side of the consensus trade, which seems bad if the price continues to rise.

I understand that market makers can widen their bid–ask spreads to reduce flow, but doing so also makes them less competitive.

1. How do market makers hedge themsleves during these events to remain neutral during events like earnings releases or major news? At least with earnings, they can pre-empt to some extent, but what about intraday news...
2. How does this approach differ between HFTs and investment bank trading desks?
3. Market makers are required to continuously quote both a bid and an ask - but if they want to avoid trading altogether, is it acceptable (or common) for a hft to quote extremely wide spreads as a workaround?

To me, IB trading desks in particular are always vulnerable to adverse selection when trading with hf clients. I still can't wrap my head around how they survive when literally everyone else wants to do the same thing.

I’ve spent a decent amount of time digging into prediction market arb and the biggest takeaway is that most of it isn’t actually arbitrage it’s just relative value with extra steps.

Step 1: Know What You’re Doing

Buying 58 here and selling 65 there sounds like free money.
In reality, you’re usually just betting that prices converge.
That’s not true arbitrage.

Step 2: Not Every Price Difference Is Wrong

Prices differ because of:

• Liquidity
• User base
• Reaction speed to news

On platforms like Novig (no vig, more market driven), this becomes obvious fast. Some gaps are just supply and demand, not mistakes.

Step 3: Execution Is Everything

Finding a spread is easy. Executing it is not.

What actually matters:

• Slippage – kills your edge
• Liquidity – can you get size down?
• Speed – if you’re late, it’s gone

The spread you see ≠ the spread you get.

Step 4: Where the Real Edge Is

The best opportunities usually aren’t clean arb.

They’re:

• Overreactions
• Bad multi-outcome pricing
• Late-market inefficiencies

Basically spots where the market is wrong, not just different.

Reality Check

• Most arb is gone before you see it
• Execution > identification
• You’re usually trading value, not locking profit

Final Thought

The shift for me was simple:

Stop looking for arbitrage, start looking for mistakes.

I recommend investopedia for understanding terms.

Liquidity is a term used in markets to describe buyers and sellers. The more buyers and sellers, the more liquid a market is.

Buyers generally enter at a perceived low, or discounted price. To protect capital, they will place sell stop orders at a lower price (sell side liquidity).

Short sellers will enter at perceived high or premium price. They place buy stop orders at a higher price to protect capital (buy side liquidity)

Late arrivers (FOMO price chasers) also have these orders placed at these areas and become bag holders.

Support areas have resting sell orders below.
Resistance areas have resting buy orders above

I've been researching prediction markets recently and am curious about the market-making side of things.

For those who have experience providing liquidity on prediction markets, sports exchanges, or similar binary-event markets:

• How did you get started?
• What level of capital is typically needed to make market making worthwhile?
• Are most market makers operating manually, semi-automated, or fully automated strategies?
• What kind of edge is required to be profitable after fees and adverse selection?
• What infrastructure (APIs, bots, pricing models, risk systems) is considered essential?
• Are there opportunities for smaller independent operators, or is the space dominated by firms?

I'd appreciate hearing about both the technical and business side of becoming a liquidity provider. Any resources, lessons learned, or common mistakes to avoid would be helpful.

Thanks!

ive most stuck to kalshi and polymarket but im seeing more and more. it seems like the sportsbooks are also starting to adopt some type of prediction market as well. some are more sports centered like prophetx and novig but i feel like theres so many now and its hard to keep up

Let's just look at the left side of the picture and imagine you are a large player looking to buy. Ideally you would be looking for lots of sellers right? Otherwise the demand you create from buying when there aren't many sellers at a given price means you will have to buy more of your large position at a higher and higher price.

If you understand that, looking at the consolidation above the "SSL" marker, you know there are both buyers and sellers who have built positions there (and likely a higher volume of them if this is a time chart, or definitely higher volume if this is a volume chart).

So when price goes well below that "SSL" consolidation area, many buyers from that consolidation who bet on price moving up will have sell stops to get out of their positions somewhere below the "SSL" consolidation (the more conservative might even have the stop losses right on the swing lows). As their positions sell to close, they become the liquidity for the large player looking to buy.

Not only that, breakout traders might see this as a chance to short the market, so they are also induced to sell.

As you can see price consolidates just below the prior "SSL" consolidation because you had roughly equal supply (sellers) and demand (buyers), and the large buyer was likely able to fill most or all of their position at a much better (lower) price using the liquidity from sell stops and breakout traders shorting the market.

Kinda stressing about this: how liquid are sports prediction markets in reality, not the marketing blurb? im the type who wants to open a position midweek and then bail if injury news drops, but i'm worried i'll be stuck holding right up to kickoff with no takers.

i've been eyeing FanDuel Predicts because it's in my state and supposedly you can trade in/out pregame, but i have no sense of how thick the book gets 30-60 mins before kickoff.

For folks who actually trade these, what are you seeing on spread/slippage and partial fills close to game time, do limit orders hit quickly or do you end up price-chasing through a thin ladder? Also curious if you hedge by setting exits earlier in the day or if there's a reliable rush of liquidity right before lineups are announced; trying to figure out whether this is workable with smallish stakes or if i'm overthinking it.

If you can't bet on the edge, it's the same as it not existing, so of course you have to factor that in.

I’d treat liquidity as part of the edge, not a separate afterthought.

Top-of-book EV can be pretty fake on prediction markets. If fair is 55% and the screen shows 50%, that looks great, but if there’s only $20 available at 50 and the next $500 is at 53-54, your real average entry is nowhere near the headline price.

The way I’d think about it:

- calculate EV at your actual fill size, not just best displayed price

- include spread / fees / slippage

- assume exit liquidity matters if you may need to unwind

- separate “can buy $25 of edge” from “can deploy real bankroll here”

- track CLV on the filled average price, not the first tick you saw

For sports prediction markets specifically, I’d be more conservative than with normal sportsbooks because the counterparty matters. If you’re getting filled, it may be because someone disagrees with your fair, or because you’re the stale side after news moved.

So I wouldn’t just filter manually. I’d have a hard liquidity/depth rule before something even counts as +EV. Something like: minimum size available, max spread, max slippage at target stake, and still +EV after average fill.

Disclosure: I’m building The Lineup, and this is why I think prediction markets need to be treated separately from sportsbook EV. Same fair-price idea, but completely different execution/liquidity problem.

I'll tackle what I can from your questions.

One of the things to realize about market making on polymarket, is that a lot of the market making bots on polymarket are actually making much more from farming liquidity rewards (maker rebates) than they are from their actual market making efforts. That's one thing that I've learned and it really shifted the way that I think about market making bot development. I've primarily been focused on developing bots based around different forms of spread capture, such as arbitrage on the crypto up/down markets...and these markets are so incredibly efficient that I started to do trader analysis reports on some of the leading bots and realized the whole rewards farming game.

In terms of infrastructure, if you don't already know speed is your #1 priority. At the very minimum you're going to need a VPS, a polygon RPC and a bot built with Rust (C++ or python also works, but I've heard more people report Rust is faster than both...do your own research but Rust is a safe bet).

There are both, independently ran bots and institutional operators. Polymarket recently has done a lot of updates, including releasing Clob v2, their central limit order book. Clob v2 has a very rough start but it finally feels settled and I've noticed a significant increase (or decrease actually) in latency for the better. It feels more stable...don't get me wrong it has their times where it's buggy but It feels like the playing field is evened a bit in terms of ability to compete.

My last note is this...and this is something I've been trying to push more as an opinion to everyone I interact with in terms of Polymarket bot building.....look further than just the crypto up down markets....explore sports, explore esports...export daily financial markets...don't just get tunnel vision and stuck on the crypto up/down markets. They are the most efficient, competitive markets on the platform and if you're looking to develop a market making bot, or any type of bot...I very much encourage you to keep an open mind and try other categories than just those.

Good luck! feel free to DM me if you have any questions or anything

You will lose money. Dont fall for it.

i've been looking at sports prediction markets from a +EV angle, but one thing that seems easy to underestimate is liquidity
a price can look mispriced at small size, but once you account for depth, spread, and getting out later, the edge can shrink pretty fast
for people modeling or trading these markets, do you treat liquidity as part of the edge calculation, or mostly filter markets manually?

Quick version up front: prediction markets are 1x basically everywhere (Polymarket, Kalshi, Hyperliquid all included), so the leverage doesn't come from the prediction market itself - it comes from a margin layer on top. Predmart is the best margin account for prediction markets so far. Happy to be corrected on any of it.

A prediction market outcome trades as a share between $0 and $1 that moves with the probability - a contract at $0.60 means roughly a 60% chance, settles at $1 if it happens, $0 if it doesn't.

Leverage means controlling a bigger position than your cash allows - put up a fraction, borrow the rest, gains and losses run on the full amount. At 5x a 20% move in the contract is roughly a 100% move on your money, both ways. Push too far the wrong way and you get liquidated, position closed automatically to repay the loan.

Why it matters here: these markets reward conviction. If you've done the work and think a contract is mispriced, capturing that edge with unleveraged cash ties up a lot of money for a thin return. Leverage lets you size the read up without depositing more.

This is the part most coverage gets wrong. None of the major venues offer native leverage on event outcomes. Polymarket is 1x on its event contracts. Kalshi is 1x. Hyperliquid's outcome markets are 1x and fully collateralized by design. Put in $1,000, control $1,000 of shares, everywhere.

Some platforms launched leverage products, but don't conflate them. Polymarket's perps, for example, are leverage on prices (BTC, NVDA, gold), not on the binary event outcomes. They leverage a price, not an outcome. So if your edge is in event prediction, perps don't help.

That leaves a consistent gap across the whole category: the people with the strongest researched views are capped at the same 1x as everyone else, on exactly the trades where their edge is sharpest.

Since it isn't built into the contracts, it comes from a margin layer on top.

You post collateral (stablecoins or shares), a protocol lends against it so you've got buying power bigger than your deposit, and you open a position on the outcome you've got a read on. A loan-to-value ratio caps the borrow, a liquidation threshold marks where you get force-closed. Stay above it and the position lives.

By hand that's a multi-step loan-and-route mess. The streamlined version is one-click - enter an amount, drag a slider, position opens, with the borrowing and liquidation logic handled in the background.

Honest answer: leverage on event outcomes is a newer, narrower category than the prediction markets themselves.

The prediction markets are the venues where events trade (Polymarket, Kalshi, Hyperliquid). The leverage comes from a margin layer applied on top of one of them. Right now that means using a layer like PredMart on top of Polymarket, which is the largest and most liquid venue - and depth matters, because leverage needs liquidity to actually work.

Leverage isn't a strategy, it's an amplifier on one.

Sizing up a high-conviction mispricing - turns a small researched edge into something worth acting on.

Capital efficiency on slow markets - hold a months-out thesis without freezing your whole bankroll in it.

Event-driven plays - size up ahead of a catalyst, exit on the reprice. Riskiest one, the same catalyst can gap against you.

Across all of them it's an active-trader thing, not a passive-holder thing.

Liquidation - a sharp move can wipe your collateral and close you before the event even resolves.

Gaps - event prices move hard on a single headline, and a maxed position dies on one piece of news.

Carry - borrowing accrues interest the whole time, and contracts settle to $1 or $0 so a bad resolution zeroes a leveraged position. Plus smart-contract risk.

Defenses: use less than max leverage, leave a buffer, and only use audited non-custodial protocols.

Prediction markets are 1x almost everywhere, so leverage on outcomes comes from a margin layer on top. That's the whole category in one sentence. PredMart is the one built for it - non-custodial, audited, up to 5x in one click, currently on Polymarket. Whatever you use, check the audit, the custody model, and how it liquidates before you put real money behind it.

You're not overthinking it. The displayed price is only the headline. The tradeable price is price plus depth plus exit path.

On thin markets I'd rather be directionally less clever and mechanically safer: check the book, use limits, assume your exit is worse than entry, and avoid treating 63c as a real price if only a few dollars are sitting there.

This is one of the boring bits people skip, but it's where a lot of prediction-market P&L disappears.

lately i've been noticing that the price shown on a market doesn't always mean much if the book is thin.

like yeah something says 63c, but if you try to enter with any real size the actual fill can be pretty different. and then getting out can be even worse.

do you guys actually check the orderbook before entering, or do most people just place limits and wait?

Severely, especially in high edge markets making them about average due to lack of liquidity

Your questions are reasonable, but they are mostly late-stage operational questions. Asking about capital, programming languages, and infrastructure before identifying and falsifying an edge is starting at the wrong end.

A complete answer would be book-length. I learned this after building my own data pipeline, testing multiple models, finding look-ahead and execution artefacts, rejecting strategies that looked profitable offline, and replacing simple touch-based fills with queue-aware paper execution.

You do not need institutional capital to begin the research. You need reliable data, intellectual honesty, a falsifiable hypothesis, and a rigorous validation protocol. A normal remote server is enough for much of that work. A separate wallet with roughly $50 may be enough for the final live plumbing test, but only after the strategy and execution system have survived the questions below.

Economics

1. Where exactly does the profit come from: spread, rebates, information, inventory premium, or arbitrage?
2. Who pays this profit, and why?
3. Why have competitors not eliminated the edge?
4. Is the edge structural, temporary, or a data artifact?
5. What is the profit after fees, rebates, gas, slippage, and execution errors?
6. Does the strategy remain profitable without rebates?
7. Are rebates merely compensating for toxic inventory?
8. How does the edge change with position size?
9. Where is the capacity limit?
10. Under what conditions does the strategy stop working?

Market Mechanics
11. How do matching, FIFO priority, and tick size work?
12. How is maker versus taker determined, and is post-only guaranteed?
13. What are the minimum order size and price constraints?
14. What does displayed depth represent, and how reliable is it?
15. How are partial fills handled?
16. How do cancellations ahead of you affect queue position?
17. Where do new orders at the same price enter the queue?
18. How is the aggressor side determined?
19. When does the market open, close, or stop accepting orders?
20. How do resolution, merge, redemption, and settlement work?
21. Can a token be sold without holding inventory first?
22. How are YES and NO connected economically and operationally?

Data
23. What data is required: L1, L2, trades, orders, and resolutions?
24. Are both exchange timestamps and local receipt timestamps available?
25. Are seconds, milliseconds, microseconds, and nanoseconds handled consistently?
26. How are gaps, duplicates, and out-of-order events detected?
27. Can the order book be reconstructed from events?
28. How are trades with identical timestamps ordered?
29. How is look-ahead bias prevented?
30. Is polling time being mistaken for event time?
31. How is each token ID mapped to the correct YES or NO outcome?
32. Is raw data retained for reproducibility?
33. Did the API or data format change during the sample?
34. How does the live feed differ from the historical API?

Hypothesis
35. Can the edge be expressed as one falsifiable sentence?
36. What is the causal mechanism rather than merely a correlation?
37. At what horizon should the signal work?
38. Does the signal horizon match the trade horizon?
39. What is the baseline?
40. How are PASS, FAIL, and KILL defined in advance?
41. Which parameters are frozen before looking at OOS?
42. What result would invalidate the hypothesis?
43. Is the chosen regime merely the best post-hoc bucket?
44. Does the effect persist across time and markets?

Execution Model
45. How is queue-ahead calculated when an order is posted?
46. How much volume must trade before your order fills?
47. Is a fill confirmed by a touch, the trade tape, or an exchange event?
48. Are partial fills and repeated fills of one order modelled?
49. How are cancel and replace latencies handled?
50. Can a stale order fill before cancellation is confirmed?
51. How is adverse selection measured after a fill?
52. Is markout accidentally being used as both the fill trigger and evaluation metric?
53. Are missed fills and opportunity costs included?
54. Does the paper policy exactly match the live policy?
55. How different are optimistic and conservative execution models?
56. What paper-to-live haircut should be expected?

Statistics
57. Are calibration, OOS, and temporal forward periods separated?
58. Is OOS genuinely independent of parameter selection?
59. How is multiple-hypothesis testing handled?
60. Is bootstrap performed over independent markets or days rather than individual fills?
61. What are the confidence interval, effect size, and sample size?
62. Does one market account for most of the result?
63. Is the sign stable across temporal folds?
64. What do the mean, median, fifth percentile, and worst tail look like?
65. Does the result survive conservative fees and latency assumptions?
66. Was the policy frozen before forward testing?
67. Were any parameters changed after seeing forward results?

Inventory and Risk
68. How are net, gross, and directional exposure defined?
69. Are all resting orders included in worst-case exposure?
70. What happens if every open order fills simultaneously?
71. Is inventory capped per token, market, asset, and wallet?
72. How are YES and NO hedged, and are their fills actually synchronized?
73. What are the maximum losses per fill, market, day, and total capital?
74. What does P&L look like under an adverse resolution?
75. What happens to an unhedged leg?
76. How are positions from previous markets handled after rollover?
77. Is collateral reserved, and can the wallet run out of available balance?
78. Under what conditions must the strategy stop trading?

Live System
79. What is the source of truth: process memory or the exchange?
80. How are open orders, fills, and balances reconciled?
81. What happens when an order is accepted but the request times out?
82. How are duplicate orders prevented?
83. What happens during a disconnect or stale-book condition?
84. Are quotes removed whenever the system becomes uncertain?
85. What happens on SIGTERM, Ctrl+C, OOM, or a process-manager stop?
86. How is it verified that the kill switch removed every order?
87. Are heartbeat monitoring, stale-data thresholds, and alerts implemented?
88. Can the system recover after a restart without losing state?
89. Is there an audit log for decisions, orders, fills, cancellations, and P&L?
90. Can the reason for every submitted order be explained?

Deploying Capital
91. Does the system first operate in dry-run mode without hidden network writes?
92. Has it passed historical, OOS, forward-paper, and exchange-level testing?
93. Have the real balance scale, tick size, and minimum order size been verified?
94. Is a separate wallet with strictly limited capital being used?
95. Is there a staged ramp plan with explicit promotion criteria?
96. How does the live fill rate compare with paper?
97. How is real cancellation latency measured?
98. At what paper-to-live deviation is the strategy shut down?
99. Who observes the first launch, and how quickly can the system be stopped?
100. Is there evidence that the system can not only make money, but also stop trading safely?

Every time a market is wrong, people rush to say prediction markets "failed." I don't think that's the right way to look at it.

A market trading at 80% isn't saying an event is guaranteed to happen. It's saying that, given everything currently known, participants collectively think it has roughly an 80% chance. If the remaining 20% happens, that doesn't automatically mean the market was inefficient. Low-probability events occur all the time.

What interests me more is whether the market was well calibrated before the outcome. If you replayed the same situation 100 times with the same information available, would that event happen close to 80 times? That's a much better test than judging the market based on a single result.

I think too many people evaluate prediction markets with hindsight. Once the outcome is known, every missed signal feels obvious. Before the outcome, those signals are rarely as clear as people remember.

That's why I care less about whether a market was right and more about whether it was pricing uncertainty reasonably.