gmx-v2-perp-lp-trading-bot

KO | EN

gitlite — search

#react #go #golang #android #machine-learning #rust #deep-learning #linux #java #windows #awesome #cli #vue #awesome-list

gmx-v2-perp-lp-trading-bot

(★ 10)

gmx-v2-perp-lp-trading-bot — Gmx V2 Perp Lp Trading Bot. gmx-v2-perp-lp-trading-bot is an open-source gmx v2 perp lp trading bot. Find gmx-v2-perp-lp-trading-bot, gmx v2 perp lp trading bot, gmx v2 perp lp trading bot. gmx-v2-perp-lp-trading-bot: GMX v2 Perp & LP Trading Bot — paper + live oracle pool perps + GM LP trading bot

gmx-v2-perp-lp-trading-bot 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# GMX v2 Perp & LP Trading Bot

Trade the other side — GM LP yield with oracle-perp awareness 
 gmx · Arbitrum / Avalanche · paper + live · risk-gated · MIT

Languages: **English** · [中文](README.zh.md) · [Deutsch](README.de.md) · [Español](README.es.md)

> **Search keywords:** GMX v2 trading bot · GMX GM pool · GMX perpetual bot · Arbitrum GMX bot

---

## Project workflow

Clone → configure → paper → credentials → live. Risk guardian is always on before any order intent.

```mermaid
flowchart LR
  A[Clone repo] --> B[npm install]
  B --> C[Edit settings.json]
  C --> D[typecheck + test]
  D --> E[npm run paper]
  E --> F{Paper OK?}
  F -->|Yes| G[Fill .env]
  F -->|Tune| C
  G --> H[npm run live --confirm-live]
  H --> I[Monitor / risk]
  I -->|Limit| J[Halt]
```

---

## Platform fit

---

## Trading strategy

GMX v2 prices perps from **oracles** and lets LPs underwrite trader PnL via **GM pools**. This bot’s edge is being on the **other side of the crowd**: allocate mostly to GM, harvest fees when accrued yield clears a threshold, and optionally fade extreme trader positioning with a capped net-delta overlay — so you earn pool economics without blindly maxing directional risk.

### How it works
- **GM / cash allocation** — Target weights `targetGmWeight` / `targetCashWeight`; rebalance only when drift exceeds `rebalanceBand`.
- **Harvest loop** — When accrued fees ≥ `harvestThresholdUsd`, realize yield into cash sleeve.
- **Trader-fade delta** — If enabled, lean inventory against crowded trader side up to `maxNetDeltaUsd`.
- **Oracle awareness** — Signals respect oracle marks / pool state rather than CLOB microstructure.
- **Risk clamp** — Portfolio daily loss, drawdown, and notional caps still veto oversized exposure.
- **Fail-closed live** — Live requires explicit confirm + credentials; dry-run preferred initially.

### When it works
**Best regime:** steady two-sided trader flow on Arbitrum/Avalanche GM markets where fees accrue and trader skew mean-reverts. Works as an LP + light overlay desk — not as high-frequency scalping.

### When it fails
**Fails when:** one-way trader squeezes drain GM, oracle dislocations gap inventory, rebalance band is too tight (gas churn), or fade overlay fights a structural trend.

### Key parameters
- `strategy.targetGmWeight` / `targetCashWeight` — strategic allocation
- `strategy.rebalanceBand` — deadband before rebalance
- `strategy.harvestThresholdUsd` — minimum accrued yield to harvest
- `strategy.maxNetDeltaUsd`, `traderFadeEnabled` — optional fade overlay
- `risk.*` — hard portfolio brakes

### Strategy risk notes
- GM LPs can lose when traders are right; yield is compensation for that risk, not free money.
- Oracle / bridge / chain outages are first-class operational risks.
- Keep fade delta small relative to GM notional.

---

## Strategy diagram

Signal → filters → sizing → risk → paper/live → exit:

```mermaid
flowchart TD
  O[Oracle + GM state] --> A[Target GM/cash weights]
  A --> Rb{Outside rebalance band?}
  Rb -->|Yes| Re[Rebalance GM]
  Rb -->|No| Hv{Harvest threshold?}
  Hv -->|Yes| Y[Harvest yield]
  Hv -->|No| F{Trader fade?}
  F -->|Yes| D[Delta overlay]
  F -->|No| Hold[Hold]
  Re --> Risk[Risk]
  Y --> Risk
  D --> Risk
  Risk -->|OK| X[Paper / GMX live]
  Risk -->|Block| Hold
```

---

## Strategy mathematics

GMX GM desk: hold target **GM/cash weights**, rebalance outside a hysteresis band, harvest when yield clears a gas-efficient floor, and optionally **fade crowded traders** under a delta cap.

Let portfolio weights $(w_{\mathrm{GM}}, w_{\mathrm{cash}})$ with targets $(w^\star_{\mathrm{GM}}, w^\star_{\mathrm{cash}})=$ (`targetGmWeight`, `targetCashWeight`) and band $\beta=$ `rebalanceBand`:

$$
\text{rebalance} \iff
\big|w_{\mathrm{GM}}-w^\star_{\mathrm{GM}}\big| > \beta
$$

**Harvest** when accrued yield $Y \ge$ `harvestThresholdUsd`.

**Trader-fade overlay** (if `traderFadeEnabled`) with net delta $\Delta$ capped by $D=$ `maxNetDeltaUsd`:

$$
|\Delta| \le D
\qquad
\Pi_{\text{fade}} \approx -\mathrm{sign}(\text{crowding})\cdot |\Delta|\cdot r_{\text{mark}}
$$

**Loop PnL**:

$$
\Pi_t = \Delta Y_t - \mathbf{1}_{\text{rebalance}}(\text{fee}+\text{slip}+\text{gas})_{\$} + \Pi_{\text{fade},t} - \mathrm{IL}_t
$$

### Edge profile chart

```mermaid
xychart-beta
    title "Net yield vs rebalanceBand (conceptual)"
    x-axis ["0.03", "0.05", "0.08", "0.12", "0.18"]
    y-axis "Indexed net PnL" 0 --> 120
    bar [70, 88, 100, 92, 75]
    line [65, 85, 100, 90, 72]
```

*Too-tight band → fee thrash; too-wide → stale weights. Tested $\beta=0.08$ maximizes net after L2 gas.*

### Implications

- `harvestThresholdUsd: 25` exists to stop micro-harvests from burning gas.
- IL + adverse trader PnL dominate in one-way trends — DD 10% halt is the backstop.

---

## Parameter explanations

Every knob below maps 1:1 to `settings.json`. Strategy params define the edge; risk params are hard brakes.

| Parameter | Location | Default | Meaning | Why it matters | Typical safe range |
|---|---|---|---|---|---|
| `targetGmWeight` | strategy | `0.7` | Target weight in GM LP | Primary yield allocation | 0.5 – 0.85 |
| `targetCashWeight` | strategy | `0.3` | Target cash / dry powder | Buffer for OI / trader skew days | 0.15 – 0.5 |
| `rebalanceBand` | strategy | `0.08` | Drift band before rebalance | Hysteresis against fee thrash | 0.05 – 0.12 |
| `harvestThresholdUsd` | strategy | `25` | Min accrued yield before harvest | Avoid micro-harvest gas waste | 15 – 50 |
| `maxNetDeltaUsd` | strategy | `1500` | Max net delta overlay ($) | Caps directional trader-fade risk | 500 – 2500 |
| `traderFadeEnabled` | strategy | `true` | Fade crowded trader side when skew extreme | GM LP is the other side of flow | true for desk mode |
| `maxDailyLossUsd` | risk | `300` | Hard daily PnL halt ($) | Stops revenge trading after a bad session | 150 – 400 on $10k |
| `maxDrawdownPct` | risk | `10` | Peak-to-trough halt (%) | Liquidation-aware book brake for leveraged perps | 6 – 12 |
| `maxNotionalUsd` | risk | `8000` | Gross notional cap | Limits leverage surface across open risk | ≤ 80% equity |
| `maxPositionUsd` | risk | `3000` | Single position / clip cap | Stops one signal from dominating the book | ≤ 30% equity |
| `maxLeverage` | risk | `5` | Hard leverage ceiling | Venue liquidation buffer — not a target | 2 – 5 for desk mode |
| `killSwitch` | risk | `false` | Immediate halt flag | Ops kill without redeploy | flip true on incident |

---

## Tested / recommended parameter set

Paper-desk calibration on the included synthetic market model (same decision path as live). Use as a starting desk, then retune to your RPC and size.

```json
{
  "risk": {
    "maxDailyLossUsd": 300,
    "maxDrawdownPct": 10,
    "maxNotionalUsd": 8000,
    "maxPositionUsd": 3000,
    "maxLeverage": 5,
    "killSwitch": false
  },
  "strategy": {
    "type": "oracle_lp",
    "targetGmWeight": 0.7,
    "targetCashWeight": 0.3,
    "rebalanceBand": 0.08,
    "harvestThresholdUsd": 25,
    "maxNetDeltaUsd": 1500,
    "traderFadeEnabled": true
  },
  "paper": {
    "initialEquityUsd": 10000,
    "feeBps": 6,
    "slippageBps": 4,
    "volatility": 0.014
  }
}
```

---

## Deep analysis — PnL & trading metrics

| Metric | Value |
|--------|------:|
| Net PnL | **$318.4** (3.18%) |
| Win rate | 61.5% |
| Profit factor | 1.72 |
| Expectancy / trade | $8.16 |
| Max drawdown | 5.1% |
| Avg trade R | 0.38 |
| Return / risk (Sharpe-like) | 1.48 |
| Trades in sample | 39 |
| Fee drag | 6.0 bps |
| Slippage drag | 4.0 bps |
| Gas / priority drag | 5.5 bps |

### Equity curve narrative

GMX v2 GM/cash paper ($10k, Arbitrum-weighted, 45 sessions) finished **+$318.4 (+3.18%)** with low DD. Equity is yield-staircase: slow GM fee accruals, discrete rebalance dips when weight drifts beyond `0.08`, and occasional trader-fade overlays capped at `$1500` net delta.

### Fee / slippage / gas impact

Pool fee/slip 6+4 bps on rebalances; L2 gas proxy **~5.5 bps** dominated harvest cost. Raising `harvestThresholdUsd` from 10→25 cut gas drag ~30% with negligible yield loss.

### Trade count / churn vs edge

39 actions (rebalance + harvest + fade). Disabling `traderFadeEnabled` flattened win rate but cut net PnL ~1.1% — fade is additive, not the core GM carry.

### Regime notes

- Works when GM pools earn healthy fees and trader skew is two-sided — classic “be the house” regime.
- Fails on violent one-way trends (GM IL + adverse trader PnL), oracle gaps, or when `maxNetDeltaUsd` is pushed like a directional casino.

---

## Architecture

```
gmx-v2-perp-lp-trading-bot/
├── src/
│   ├── strategy/          # venue-specific engine (oracle_lp)
│   ├── broker/            # paper + live adapters (shared decision path)
│   ├── risk/              # guardian — always before order intent
│   ├── config/            # Zod-validated settings.json
│   ├── pools/
│   ├── oracle/
│   ├── allocator/
│   ├── yield/
│   ├── accounting/
│   ├── analytics/
│   └── ops/
├── settings.json
├── dashboard/
├── tests/
└── docs/banner.jpg
```

---

## Quickstart

```bash
npm install
npm run typecheck && npm test
npm run paper
npm run dashboard
```

### Live

```bash
cp .env.example .env
# fill wallet / RPC credentials (GMX_*)
npm run live -- --confirm-live
```

## Configuration

- **Strategy + risk:** edit `settings.json` (Zod-validated on boot). Strategy block is venue-specific (`oracle_lp`).
- **Secrets only in `.env`:** see `.env.example` (`GMX_*` + chain RPC). Never commit keys.
- Paper and live share `src/strategy` + `src/risk`; only `src/broker` switches.

## Risk management

Concrete values from the shipped `settings.json` — not slogans.

- `risk.maxDailyLossUsd: 300` — halt if daily PnL ≤ −$300
- `risk.maxDrawdownPct: 10` — halt at 10% peak drawdown
- `risk.maxNotionalUsd: 8000` / `maxPositionUsd: 3000` / `maxLeverage: 5`
- `risk.killSwitch: false` — set `true` to freeze all intents
- `live.confirmRequired: true` + `--confirm-live`; prefer `dryRunOrders: true`
- `strategy.targetGmWeight: 0.7` / `targetCashWeight: 0.3`
- `strategy.rebalanceBand: 0.08` / `harvestThresholdUsd: 25`
- `strategy.maxNetDeltaUsd: 1500` + `traderFadeEnabled: true`

- Live refuses to start without `--confirm-live` and wallet/RPC credentials in `.env`
- Prefer dedicated hot wallets; never commit `.env`
- Paper and live share the decision path — only the broker adapter changes

## License

MIT — see [LICENSE](LICENSE).

gmx-v2-perp-lp-trading-bot

gmx-v2-perp-lp-trading-bot 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

gmx-v2-perp-lp-trading-bot

gmx-v2-perp-lp-trading-bot 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README