How to Execute a Large Position with TWAP and Iceberg Orders

Work a large position into a thin Polymarket order book using TWAP and Iceberg without moving the price against yourself or showing your hand.

Overview

Prediction market books are thin. That is not a defect, it is what a market for a specific question about a specific event looks like: a few hundred shares at the touch, a few hundred more a couple of cents away, then a gap.

Which means a market order for size does not fill at the price you saw. It walks the book, and you pay the average of every level it eats. On top of that, everyone watching sees a large aggressive order and adjusts before your next one.

Two tools address this from opposite directions. This guide covers when each is right, how to tune them, and how to combine them when you need both fill certainty and a decent price.

Read time: about 15 minutes. Assumes you have placed an order before. Start with Place an order if not.

TL;DR

• TWAP splits a budget into clips over a time window. Takes liquidity, high fill certainty, costs you the spread.
• Iceberg rests a large order while showing one clip at a time. Provides liquidity, earns the spread, may not fill at all.
• Iceberg slices are post-only. Price a buy above the best bid and it is rejected for crossing the book.
• Both return a record_id and live on GET /v2/trade/strategies, not open orders.
• cancel-all does not stop them. Cancel strategies individually.
• Both keep running if your process dies, which is a feature, not a risk to manage away.

What you will do

• Work out whether your order is actually large relative to a given book
• Run a TWAP with tuning you can justify rather than defaults you copied
• Rest an Iceberg without tripping the post-only rule
• Monitor both on the right endpoint
• Combine them so a deadline is met without paying the spread on the whole position
• Cancel safely, understanding what happens to size already filled

What you will need

Knowledge

• Familiarity with order books and the difference between taking and providing liquidity
• Place an order covers the basics if you need them

Tools and access

• A Bravado API key with trade.execute and trade.cancel
• Enough collateral for the full position

export BRAVADO_API_KEY="your-bearer-token"
export BASE="https://bravado-api-k7kaq.ondigitalocean.app"

Is your order actually large?

“Large” is relative to the book, not to your account. A useful rule: if your order is more than about a quarter of the size resting within a cent or two of the touch, you will move the price.

Consider a book with the best ask at 0.62 for 200 shares, 0.64 for 300, then 0.68.

Your order	What happens	Effective price
150 shares	Fills at the touch	0.62
500 shares	Eats two levels	~0.633
1,000 shares	Eats three, moves the market	~0.66

That last row is a 6% worse entry than the price you were looking at, on a market where 6% is a meaningful edge. That is the cost this guide exists to avoid.

The two approaches

	TWAP	Iceberg
Splits by	Time	Displayed size
Liquidity	Takes	Provides
Fill certainty	High, it keeps buying	Low, waits to be hit
Spread	You pay it	You earn it
Best when	You must be filled by a deadline	You would rather wait than overpay
Visibility	Repeated small takes	One small resting clip

The choice is really about whether time or price is your binding constraint.

TWAP: spread over a window

curl -X POST $BASE/v2/trade/order \
  -H "Authorization: Bearer $BRAVADO_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "type": "TWAP",
    "symbol": "71321045679252212594626385532706912750332728571942532289631379312455583992646",
    "side": "BUY",
    "budget_usdc": "2000",
    "execution": {
      "duration_sec": 3600,
      "interval_sec": 60,
      "price_tolerance_pct": 2,
      "randomize_pct": 10
    }
  }'

{ "record_id": "rec_twap_abc123", "status": "ACTIVE", "type": "TWAP" }

$2,000 over an hour, in roughly 60 clips of about $33 each.

Tuning it

duration_sec and interval_sec: how many clips

Both in seconds, and interval_sec has a minimum of 10. Clip count is roughly duration_sec / interval_sec, which sets clip size.

Too few clips and each one is large enough to move the market, defeating the purpose. Too many and the strategy outlives the edge you were trading on. A useful sanity check: clip size should sit comfortably inside the size resting at the touch.

Also check the floor. A market order needs $1 minimum, so budget_usdc / clip_count must stay above that. $50 over 100 clips is 50 cents a clip and will fail.

price_tolerance_pct: your spike protection

A percent from 0 to 100. Clips are skipped when price has moved beyond this from the strategy’s reference.

This is what stops a TWAP buying into a spike. Set it too tight and the strategy quietly does nothing in a moving market, then you discover at the end that almost nothing filled. Set it too loose and it fills through news you would rather have waited out. Around 2% is a reasonable starting point on a liquid market.

randomize_pct: hiding the pattern

A percent that jitters clip timing and size. Without it, execution is a metronome: same size, same interval, entirely predictable to anyone watching the tape. 10 is enough to break the pattern without meaningfully changing the average.

Iceberg: show a slice, hide the rest

curl -X POST $BASE/v2/trade/order \
  -H "Authorization: Bearer $BRAVADO_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "type": "ICEBERG",
    "symbol": "...",
    "side": "BUY",
    "price": "0.60",
    "size": "5000",
    "execution": { "clip_size": "100" }
  }'

The book shows 100 shares at 0.60. As that clip fills, another replaces it, until all 5,000 are done or you cancel.

Slices are post-only. For a buy, price must be at or below the current best bid. Price it above and the slice is rejected with order crosses book, because it would take liquidity rather than provide it.

Iceberg cannot be used to fill aggressively. That is not a limitation to work around, it is the entire mechanism: you are paid the spread precisely because you wait.

Two constraints on clip_size:

• At least 5 shares, the venue minimum for a resting order.
• Small enough not to signal size, large enough that refills are not constant. Somewhere near the typical resting size at the touch usually works.

Monitor on the right endpoint

Both strategies live here, and not on open orders:

curl $BASE/v2/trade/strategies \
  -H "Authorization: Bearer $BRAVADO_API_KEY"

{
  "strategies": [
    {
      "record_id": "rec_twap_abc123",
      "type": "TWAP",
      "status": "ACTIVE",
      "symbol": "713210456…",
      "filled_size": "1204.301075",
      "remaining_size": "1795.698925",
      "average_price": "0.6215"
    }
  ]
}

An Iceberg appears in two places at once: the parent as a record_id here, and whichever slice is currently resting as a child order_id on /v2/trade/orders/open. That is expected, not a duplicate.

PENDING means accepted and waiting for entry conditions, not rejected. Cancelling and re-placing on PENDING churns fees and prevents the strategy from ever working.

Cancel safely

curl -X DELETE $BASE/v2/trade/strategies/rec_twap_abc123 \
  -H "Authorization: Bearer $BRAVADO_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)"

Cancelling stops future clips. Everything already filled is a position you now hold, so read filled_size before deciding what to do next. Cancelling a half-executed TWAP leaves you with half a position, which may or may not be what you want.

POST /v2/trade/orders/cancel-all clears CLOB orders only. A TWAP keeps running afterwards. If your “flatten everything” path is wired to that endpoint alone, you are not flat.

A correct flatten:

write("POST", "/v2/trade/orders/cancel-all")

for s in read("/v2/trade/strategies")["strategies"]:
    if s["status"] in ("ACTIVE", "PENDING"):
        write("DELETE", f"/v2/trade/strategies/{s['record_id']}")

Combining both

When you want most of the position cheaply but cannot risk being unfilled:

1. Rest 70% as an Iceberg at or below the touch

   If the market comes to you, this is the cheap fill and you earn the spread on it.

2. Run a TWAP for the other 30% on your deadline

   Sized so that even if the Iceberg never fills, you end up with a position you can live with.

3. Poll filled size across both records

   Do not assume either completes. Track filled_size on each.

4. Cancel the Iceberg when the TWAP finishes

   Otherwise it keeps resting and you accumulate more than you intended.

That last step is the one people forget, and it is the expensive one: a forgotten Iceberg quietly builds a position long after the strategy that justified it has ended.

Venue constraints

Rejections that trace to these are venue rules, not Bravado validation:

• Resting orders need at least 5 shares, which floors clip_size.
• Market orders need $1 notional, which floors TWAP clip size.
• Prices are decimal probabilities from 0.001 to 0.999.
• Iceberg slices must be passive.

Wrapping up

TWAP buys time-weighted certainty and pays the spread for it. Iceberg earns the spread and pays for it in fill risk. Neither is better; they solve different constraints, and combining them lets you choose the ratio.

The operational detail that matters most: both run on Bravado’s side, so they survive your process dying. That is what makes them different from a scheduler you write yourself, which stops mid-position the moment it is interrupted.

Frequently asked questions

My iceberg slice was rejected for crossing the book.

The slice price was above the best bid on a buy, which would take liquidity. Slices are post-only. Lower the price to at or below the touch.

My TWAP shows almost nothing filled after an hour.

Most likely price_tolerance_pct is too tight and the market moved outside it, so clips were skipped. Check filled_size against elapsed time and widen the tolerance if the market is genuinely trending.

Why does my iceberg appear twice?

The parent strategy is on /v2/trade/strategies and the currently resting slice is on /v2/trade/orders/open. One order, two representations.

I cancelled everything but a strategy kept executing.

cancel-all covers CLOB orders only. Cancel each strategy individually with DELETE /v2/trade/strategies/{id}.

Can I change duration or clip size on a running strategy?

Cancel and re-place with the parameters you want, and account for what already filled when sizing the replacement.

What happens if the market resolves mid-execution?

Resolution settles outcome tokens at $1 or $0, and an unfilled strategy will not fill afterwards. Review running strategies as an event approaches. See UMA resolution.

Resources

• Place an order, all eight order types
• Track order and strategy state, where strategies live
• Build a trading bot, choosing order type by size programmatically
• Quote both sides, the market making alternative
• Trade API reference, full field documentation
• Error reference, rejection messages and causes