Planemo Intern

@PlanemoTrading
algorithmic trading platform for perpetual futures. alpha live on extended, grvt & lighter.

[Planemo Trading @PlanemoTrading]

We launched yesterday on Nado.

Just queried builder stats via their official API.

Happy to report that we have been the #1 builder over the past 24 hours.

Wen official builder stats dashboard @nadoHQ?

[Planemo Trading @PlanemoTrading]

gNado!

We are live on @nadoHQ.

Today we are launching our entire suite of trading algorithms on Nado.

This also marks the first time we are making our statistical arbitrage system publicly available across exchanges.

This has been a long time in the making, all details in the thread below.👇

[Planemo Trading @PlanemoTrading]

We at @PlanemoTrading were among the first 15 users to implement priority fees on @HyperliquidX for internal testing and are a top10 wallet by fees spent so far.

From our own tests across 1,000+ transactions, we can see that this update serves effectively as a speedbump for IOC orders on HIP3.

Our latency increased by roughly the 8x45ms that is outlined in the docs as maximum possible speed increase.

It would be interesting to analyze the current user base of priority fees to find out what strategies are being run that justify implementation.

The docs mention the reasoning behind priority fees is to prevent the escalation of an arms race of ultra-fast market makers.

We are not ultra-fast nor a market maker.

We so far don't see the value of paying 8bps extra just to get the speed we got "for free" a week ago.

Our strategy does not necessarily benefit from being faster than other takers, but it certainly hurts performance if we are slower while order cancel/refresh stays the same as pre-update.

So looks like some strategies could end up as collateral damage here as an unintended (?) side effect.

[Planemo Trading @PlanemoTrading]

Last 40 or so days of one of our stat arb users.

[Planemo Trading @PlanemoTrading]

Fun experiment on @extendedapp with interesting results:

We let out Surge Pro algo run for 1 week (epoch 48) in combination with having another single open position all week. Stats were $2M in taker (aka more points) volume, with $20k in constant open interest. Costs were $300.

Then last week (epoch 49), we deactivated Surge Pro and just kept the open position as is. Stats were just the $20k constant open interest. Costs 0.

You can see the results in the screenshots attached, but the amount of distributed points differ by a factor of 80!

422 points vs 5 points.

It's been all the rage on Twitter the last few weeks to proclaim that open interest, holding time and whatnot is the number 1 factor for points, but the numbers tell a different story.

As long as Extended launches above $150M FDV, it's hard to imagine a scenario in which you lose money with a cost per point basis significantly below $1.

A few users have recently done the math and have re-started using our Surge algos for farming Extended.

That's also why we are consistently ranked #1 there in terms of volume and fees generated for Extended (more points).

Give it a try: https://t.co/qFnlIOJgrC

[Planemo Trading @PlanemoTrading]

just try it yourself

https://t.co/DHN6yaLtJa

[Planemo Trading @PlanemoTrading]

Toxic flow is one of the main killers of stat arb profitability.

If you want to learn about toxic flow, here's a long post about the basics and how we're building defenses against it.

(Implementation will be a quarter-long effort from our engineering team and the first version will likely go online at the end of this month.)

When you run a delta-neutral stat arb strategy, you're quoting maker orders on one venue and hedging with taker orders on another. The edge is the spread dislocation between the two.

The problem: not all fills are equal.

Some fills happen because random noise pushed the spread wide enough for your signal to trigger. These are clean fills. The edge holds through your hedge execution, you capture the dislocation, everyone's happy.

Other fills happen because an informed participant, someone who knows where price is going next, deliberately takes your resting liquidity.

They fill you because your quote is mispriced relative to information they have.

By the time your hedge leg executes 100-500ms later, the edge has evaporated. Sometimes it's gone negative.

This is toxic flow. You got picked off.

The insidious part: both fills look identical at the moment of execution. Same spread, same signal, same z-score. You can't tell them apart by looking at your entry conditions. The difference only reveals itself after the fact, when you measure how much edge survived through hedge execution.

For stat arb specifically, this is devastating because:

Your maker leg sits in the order book advertising exactly where you think fair value is. Every informed trader on the venue can see it. When they know price is about to move, your quote is the cheapest liquidity available. You become the exit door for someone else's alpha.

The faster and more informed the flow, the worse it gets. Your fill rate might look great, but your PnL doesn't match, because you're systematically filled on the wrong side of information events.

So how do you defend against it?

You need to predict whether the next fill will be toxic BEFORE you get filled. Not after. Not during. Before.

This means reading the microstructure of the order book in real time and recognizing the patterns that historically preceded bad fills. Then deciding whether to even have your quote live at that moment.

Here's what our engineering team is currently building and integrating, a full toxic flow detection and prevention framework:

Order Book Microstructure Signals

The order book is not static. It's a living structure that changes shape before major moves.

We're tracking:

Book imbalance at best bid/ask: the ratio of resting size at the top of book on each side. When you're quoting a buy and ask-side depth massively outweighs bid-side depth, that's adverse. This is the single highest-value predictor of toxic fills.

Depth-weighted imbalance across multiple levels: same concept but exponentially weighted across the top 5 levels. A thin bid side with stacked asks 2-3 ticks deep is a stronger signal than just level 1.

Imbalance rate of change: static imbalance matters less than accelerating imbalance. If bid depth is draining over the last 200ms to 1 second, that's a momentum signal the book is about to move through your level. We track the delta across three lookback windows.

Spread compression and expansion dynamics: if the spread is tightening toward your quote, more participants are competing. Tight spreads on Hyperliquid tend to precede directional moves. We track spread in basis points and its short-term velocity.

Order arrival and cancellation velocity: counting new orders placed and cancelled per unit time on each side. A spike in cancellations on your side of the book is a strong adverse selection signal. Other makers are pulling their quotes. They know something.

Queue position erosion: if orders ahead of you in the queue are being cancelled (not filled), your effective queue position improves. Paradoxically this is dangerous. Informed makers are withdrawing, leaving your quote as the last available liquidity.

Trade Flow Signals

Beyond the static book, the stream of actual trades tells a story:

Recent trade direction and intensity: classifying trades as buyer-initiated or seller-initiated and computing signed volume over trailing windows of 1, 5, 15, and 60 seconds. Heavy one-directional aggression preceding your quote is the classic toxic setup.

Trade size distribution analysis: informed flow tends to come in specific size clusters. We profile historical toxic fills by aggressor order size and flag when incoming sizes exceed 1.5 standard deviations above the rolling mean.

Trade arrival rate deviation: under normal conditions, trade arrivals approximate a Poisson process. When inter-arrival time drops significantly below the expected rate - a burst of trades - it often indicates informed flow or a liquidation cascade.

Volume-weighted price momentum: short-term VWAP slope over recent trades. If VWAP is trending against your quote direction, the edge is decaying before you even get filled.

Aggressor repetition tracking: if the same address is repeatedly taking liquidity on one side in quick succession, that's likely an informed participant. Concentrated flow from single wallets gets flagged.

Cross-Venue and Cross-Provider Signals

For stat arb, the relationship between venues is critical:

Price dislocation velocity: we already compute cross-provider spreads. But the speed at which dislocations close tells you about information content. If the dislocation you're arbing is closing rapidly before you get filled, the move is information-driven, not noise. We track the half-life of recent dislocations per pair.

Lead-lag relationship shifts: identifying which provider currently leads price discovery. If the venue where you're quoting maker is lagging, your fill is more likely toxic because you're being picked off by someone arbing the same dislocation faster than you.

Funding rate and its derivative: on Hyperliquid perps, funding rate changes reflect positioning imbalances. Rapidly shifting funding signals directional pressure building against your quote direction.

Basis dynamics: tracking basis shifts between the perp you're quoting and correlated instruments. Accelerating basis movement adverse to your quote direction is informed flow repricing the market.

Correlated asset moves: BTC and ETH move first. Alts follow. We track real-time price changes in major assets as leading indicators, computing beta-adjusted expected moves for the asset we're quoting.

Volatility Regime Detection

High-volatility environments have structurally more toxic flow because price moves faster than your hedge leg can execute:

Realized volatility regime classification: rolling realized vol over 1-minute, 5-minute, and 15-minute windows. Above certain thresholds, we widen edge requirements or stop quoting entirely.

Volatility of volatility: sudden vol spikes are more dangerous than sustained high vol, because the model is calibrated to the prior regime. We track if realized vol just jumped more than 2 standard deviations from its own rolling mean.

Implied volatility proxy: using the width of market maker quotes as a proxy for expected vol. When spreads across the market suddenly widen, it signals a regime change.

Fill-Specific Forensics

Your own fill history is training data:

Time-to-fill analysis: orders that get filled very quickly after being placed are more likely toxic. Immediate fills mean someone was waiting to pick you off. Orders that rest before filling tend to be noise flow. We build Bayesian priors from the distribution of fill times for toxic vs clean fills.

Fill size relative to quote size: full instant fills are more toxic than partial fills. Full fills suggest the aggressor specifically wanted your liquidity at your price, implying they believe it's mispriced.

Position in aggressor sweep: if your fill was part of a larger sweep across multiple price levels, you were the first stop on a directional move.

Timing and Temporal Signals

Time since last major price move: mean reversion vs momentum detection. If the dislocation appeared after a sharp move that's less than T seconds old, there's higher probability it continues. If it's been stable, it's more likely noise-driven and safe.

Liquidation cascade detection: liquidations create forced directional flow. Sharp price move combined with high volume and specific size patterns. Avoid quoting into cascades.

Time-of-day effects: certain hours have structurally different flow toxicity. US market open, Asian market open, major macro events. Profiling historical toxic fill rate by hour and day.

Event proximity: CPI, FOMC, token unlocks. Flow becomes dramatically more toxic around scheduled events. Widening thresholds or stopping quoting within proximity of known events.

Hyperliquid-Specific Intelligence

Open interest changes: rapid OI increases suggest new directional positions being opened. OI decreases suggest closing, potentially mean-reverting. Tracking OI delta per unit time.

Known address classification: building a database that classifies Hyperliquid addresses we've been filled by as toxic or non-toxic based on historical outcomes. Over time this becomes the most accurate signal. You literally know which counterparties are informed.

The Model Architecture

All of these signals feed into a feature vector updated on every order book change. Every fill gets labeled against a pair-aware adverse-selection threshold: toxic if the realized edge at fill drops below 50% of the pair's expected bypass edge.

We train gradient-boosted trees offline on the labeled dataset with time-based splits to avoid lookahead bias.

The cost function is asymmetric: false negatives (toxic fills you didn't avoid) cost real money, false positives (clean fills you skipped) cost opportunity. We tune the threshold to minimize expected cost, not accuracy.

The goal is not to trade more. It's to stop trading at exactly the moments when the next fill will destroy your edge.

[Planemo Trading @PlanemoTrading]

imagine not needing to care about oil volatility and exploding funding rates

while farming volume and profits

[Planemo Trading @PlanemoTrading]

We might not have the most users, but maybe the happiest?

Market yesterday was pretty active = elevated volatility = big profits for our users.

[Planemo Trading @PlanemoTrading]

day 4 on hyperliquid in the books and a new ATH record of 82% of users in profit in the last 24 hours.

can you find any other builder on @HyperTracker that comes close to this?

it's usually hard to find anything above 35%...

try it out right now and farm all HIP3 @Dreamcash (+$USDT rewards), @tradexyz, @markets_xyz and @felixprotocol with our stat arb algo.

stop burning a "budget" to farm volume, start earning via positive PnL and consider points as a bonus

[Planemo Trading @PlanemoTrading]

he's not wrong

[Planemo Trading @PlanemoTrading]

day4 on @HyperliquidX so far looking very solid

we measure farming not in cost per million, but in profit per million

btw wen custom logo and info for our builder @HyperTracker

[Planemo Trading @PlanemoTrading]

Another goated trade.

This guy just farmed $80k @tradexyz volume in 5 minutes and made $45 profit on top.

Stop using inferior platforms where you lose ABSURD amounts of money per $1M/volume ($200 - 700 often) and try out our stat arb algo:

https://t.co/PoLuYbigp5

[Planemo Trading @PlanemoTrading]

Day 3 on @HyperliquidX doing stat arb on @markets_xyz @tradexyz @felixprotocol @Dreamcash

74% of users profitable today (including builder fees).

The market drop in the evening led to some great entries & exits.

Volatility = we print even more

[Planemo Trading @PlanemoTrading]

yo

[Planemo Trading @PlanemoTrading]

Get started today and farm @HyperliquidX and HIP3 with a positive PnL:

https://t.co/whFKxfFpAx

[Planemo Trading @PlanemoTrading]

2)

What happens normally is that one leg exits at a loss, while the other exits at a profit, and the net amount is positive.

But things sometimes look different, especially if you have a strong tailwind in form of great entries. Great entries often let you exit both legs at a profit, as in the example below.

[Planemo Trading @PlanemoTrading]

We've been live on Hyperliquid for 3 days now.

Time to highlight our favourite trades our stat arb algo executed for our users.

1)

By far the best one, is this crazy SILVER trade that happened on Friday.

Big dislocations on the SILVER @tradexyz and @felixprotocol market there, +$200 within a few minutes.

(We usually are happy to exit a trade with a few bps of profit). The $200 here translate to a few % on the user's collateral.

[Planemo Trading @PlanemoTrading]

Lovely.

Great entries allow for a profitable exit on both legs from time to time.

[Planemo Trading @PlanemoTrading]

CL (xyz) - BRENTOIL (xyz) statistical arbitrage is now live on @PlanemoTrading.

Farm double the volume on @tradexyz compared to our usual pairs.

https://t.co/PoLuYbigp5