Sistem Trading Algoritmik Berbasis Regime Pasar dengan Manajemen Risiko Berlapis dan Analisis Mikrostruktur pada Aset Kripto BTC/USDT (Evolusi QPS)

Pasar aset kripto, khususnya pasangan BTC/USDT, dicirikan oleh tingkat volatilitas yang ekstrim, asimetri informasi yang tinggi, serta kecepatan transisi kondisi pasar (market regime) yang sangat dinamis antara fase trending dan sideways. Karakteristik ini membuat pendekatan tradisional menggunakan indikator teknikal tunggal rentan menghasilkan sinyal perdagangan palsu ketika kondisi pasar mengalami pergantian mendadak.

The cryptocurrency asset market, particularly the BTC/USDT pair, is characterized by extreme volatility levels, high information asymmetry, and highly dynamic market regime transition speeds between trending and sideways phases. These characteristics make traditional approaches using single technical indicators prone to producing false trading signals when market conditions change abruptly.

Sistem trading algoritmik generasi awal seperti Bot A (BitTrade-v2 Engine) dirancang untuk memitigasi kelemahan tersebut melalui mekanisme deteksi market regime yang berupaya mengklasifikasikan kondisi pasar terlebih dahulu sebelum menentukan eksekusi strategi. Meskipun arsitektur awal berhasil memperkenalkan pengamanan dasar, data empiris nyata menunjukkan sebuah fenomena sistematis: mayoritas sinyal perpotongan indikator (EMA Golden Cross) terpicu akibat manipulasi likuiditas buatan atau *fake-out*, bukan oleh minat fundamental pasar sesungguhnya.

Early-generation algorithmic trading systems like Bot A (BitTrade-v2 Engine) were designed to mitigate these weaknesses through market regime detection mechanisms that attempt to classify market conditions first before determining strategy execution. Although the initial architecture successfully introduced basic safeguards, real empirical data demonstrated a systematic phenomenon: the majority of indicator crossover signals (EMA Golden Cross) were triggered by artificial liquidity manipulation or fake-outs, rather than by genuine fundamental market interest.

Bagaimana mengidentifikasi dan memisahkan momentum pergerakan harga yang asli (genuine breakout) dari pergerakan harga palsu (fake-out manipulation) secara *real-time*? Dan bagaimana cara mencegah sistem mengalami kebangkrutan (*ruin*) pada masa transisi pasar yang dipenuhi rentetan kerugian (*losing streaks*) beruntun?

How to identify and separate genuine price movement momentum (genuine breakouts) from fake price movements (fake-out manipulation) in real-time? And how to prevent the system from experiencing bankruptcy (ruin) during market transition periods filled with consecutive losing streaks?

Tujuan utama dari penulisan skripsi ini adalah merancang, mengimplementasikan, dan mengevaluasi kinerja arsitektur Quant Professional System (QPS) pada BitTrade-v2. Secara spesifik, penelitian ini memformulasikan manajemen ukuran transaksi dinamis melalui Kelly Criterion yang dimodifikasi oleh Sharpe Ratio, serta mengembangkan modul High-Frequency Microstructure Analysis melalui kalkulasi tingkat ketidakseimbangan buku pesanan (Order Book Imbalance / OBI).

The main objective of writing this thesis is to design, implement, and evaluate the performance of the Quant Professional System (QPS) architecture on BitTrade-v2. Specifically, this research formulates dynamic transaction size management through the Kelly Criterion modified by the Sharpe Ratio, and develops a High-Frequency Microstructure Analysis module through the calculation of Order Book Imbalance (OBI) levels.

Deteksi market regime merupakan fondasi dari seluruh sistem perdagangan adaptif modern. Lo & MacKinlay (1988) membuktikan bahwa pasar keuangan bersifat non-stasioner, bergerak antara fase momentum dan reversion secara bergantian. Pendekatan umum adalah memanfaatkan tingkat volatilitas persentase (standard deviation) sebagai ambang batas klasifikasi.

Market regime detection is the foundation of all modern adaptive trading systems. Lo & MacKinlay (1988) proved that financial markets are non-stationary, moving alternately between momentum and reversion phases. A common approach is to utilize percentage volatility levels (standard deviation) as a classification threshold.

Dalam teori investasi kuantitatif, Kriteria Kelly adalah formula matematika yang menentukan porsi kekayaan optimal untuk dipertaruhkan. Sistem profesional modern mengadopsi Fractional Kelly, seperti Quarter Kelly. Selain itu, alokasi modal tidak hanya memperhitungkan peluang kemenang (win rate), melainkan juga disesuaikan dengan imbal-risiko aset tersebut yang biasa direpresentasikan dengan nilai Sharpe Ratio (Pardo, 2008).

In quantitative investment theory, the Kelly Criterion is a mathematical formula that determines the optimal portion of wealth to bet. Modern professional systems adopt Fractional Kelly, such as the Quarter Kelly. In addition, capital allocation not only takes into account winning opportunities (win rate), but is also adjusted to the risk-reward of the asset, which is usually represented by the Sharpe Ratio value (Pardo, 2008).

Mikrostruktur pasar mempelajari proses pembentukan harga pada tingkat terendah: pencocokan setiap *order bid* dan *ask*. Order Book Imbalance (OBI) adalah matriks asimetri utama yang mengungkap niat tersembunyi pasar (O'Hara, 1995). Pada aset kripto, institusi pembentuk harga (market makers) menempatkan *limit order* masif untuk menciptakan Sell Wall atau Buy Wall.

Market microstructure studies the price formation process at the lowest level: the matching of each bid and ask order. Order Book Imbalance (OBI) is the primary asymmetry matrix that reveals the market's hidden intentions (O'Hara, 1995). In crypto assets, market makers place massive limit orders to create a Sell Wall or Buy Wall.

BitTrade-v2 Engine dibangun sepenuhnya menggunakan bahasa pemrograman Rust, dipilih karena jaminan keamanan memori tingkat tinggi (memory safety) dan kemampuan *multithreading* asinkron ekstrem melalui ekosistem Tokio. Persistensi data dilakukan melalui basis data PostgreSQL menggunakan sqlx.

BitTrade-v2 Engine is built entirely using the Rust programming language, chosen for its high-level memory safety guarantees and extreme asynchronous multithreading capabilities through the Tokio ecosystem. Data persistence is handled via a PostgreSQL database using sqlx.

Sistem dirancang mengalirkan setiap perubahan harga menjadi keputusan sepersekian detik melalui pipeline 5 tahap: Ingestion (Data), Heuristic (Pola), Conclude (Strategi), Risk & Validate (Manajemen Risiko), dan Executor (Eksekusi virtual & pemotongan biaya).

The system is designed to flow every price change into split-second decisions through a 5-stage pipeline: Ingestion (Data), Heuristic (Patterns), Conclude (Strategy), Risk & Validate (Risk Management), and Executor (Virtual execution & fee deduction).

Masalah krusial pada versi awal adalah hiper-akumulasi posisi yang tak terkendali seiring naiknya harga. Pada Fase 4.1, disahkan aturan "Maksimal Pyramiding 3 Layer". Di samping itu, diterapkan pula filter *whipsaw* di mana konfirmasi tren harus bertahan selama 2 menit berturut-turut.

A crucial problem in the initial version was the uncontrolled hyper-accumulation of positions as prices rose. In Phase 4.1, a "Maximum 3 Layer Pyramiding" rule was enacted. In addition, a whipsaw filter was applied where trend confirmation must last for 2 consecutive minutes.

Pengujian coba jalan sistem selama 48 jam menghasilkan rasio kemenangan 30.0% dengan Net P&L kumulatif minus -$12.22. Pembedahan kronologis mengungkap pola destruktif: FOMO Re-entry Trap (masuk lagi pasca untung di harga mahal) dan Dead Volume Fake-out (masuk saat volume kosong).

A 48-hour test run of the system yielded a 30.0% win rate with a cumulative Net P&L of minus -$12.22. Chronological dissection revealed destructive patterns: FOMO Re-entry Trap (re-entering post-profit at expensive prices) and Dead Volume Fake-out (entering when volume is empty).

Transisi dan peluncuran resmi arsitektur Quant Professional System (QPS) pada lingkungan produksi live dieksekusi secara mulus pada tanggal 12 Juli 2026. Pembaruan ini mengubah paradigma bot dari sistem reaktif menjadi prediktif berorientasi risiko.

The transition and official launch of the Quant Professional System (QPS) architecture into the live production environment was seamlessly executed on July 12, 2026. This update shifts the bot's paradigm from a reactive system to a risk-oriented predictive one.

Modul QPS Risk menyapu tabel riwayat transaksi SQL untuk menganalisis 20 transaksi SELL terakhir (Window T-20). Modul menghasilkan rata-rata pengembalian bersih per perdagangan (μ) dan volatilitas pengembalian (σ) untuk membentuk Sharpe Ratio.

The QPS Risk module sweeps the SQL transaction history table to analyze the last 20 SELL transactions (T-20 Window). The module generates the average net return per trade (μ) and return volatility (σ) to form the Sharpe Ratio.

Apabila Sharpe Ratio bernilai negatif, QPS memberikan instruksi perlindungan modal (Capital Preservation Halt) dan menekan alokasi menjadi 0%. Apabila positif, sistem mengeksekusi porsi anggaran berdasarkan proporsi Quarter Kelly Fraction (maksimal 35%).

If the Sharpe Ratio is negative, QPS issues a Capital Preservation Halt instruction and compresses allocation to 0%. If positive, the system executes a budget portion based on the Quarter Kelly Fraction proportion (maximum 35%).

Sistem memantau kuantitas volume pada antrean *Best Bid* dan *Best Ask* untuk menghitung koefisien tekanan melalui rasio Order Book Imbalance (OBI):

The system monitors the volume quantities in the Best Bid and Best Ask queues to calculate the pressure coefficient through the Order Book Imbalance (OBI) ratio:

Parameter penolakan keras ditetapkan pada skala 0.40. Jika OBI < 0.40 (artinya 60%+ dari keseluruhan volume antrean utama dikuasai oleh Sell Wall), maka sistem membatalkan perintah beli alih-alih terjun bebas ke dalam jebakan paus pasar.

The hard rejection parameter is set at a scale of 0.40. If OBI < 0.40 (meaning 60%+ of the entire main queue volume is controlled by a Sell Wall), the system cancels buy orders instead of free-falling into a market whale trap.

Transformasi komprehensif ke arsitektur QPS memberikan perubahan yang teramat mencolok pada tabiat perilaku bot.

The comprehensive transformation to the QPS architecture provided an extremely striking change in the bot's behavioral traits.

Evaluasi pasca-deployment (live) menunjukkan anomali sistemik yang signifikan. Selama 9 hari pertama beroperasi penuh, QPS mencatat 0 transaksi. Ini bukan indikasi kesuksesan perlindungan risiko, melainkan kegagalan parameter threshold statis.

Post-deployment (live) evaluation revealed a significant systemic anomaly. During the first 9 days of full operation, QPS recorded 0 transactions. This is not an indication of successful risk protection, but rather a failure of the static threshold parameters.

Data riil membuktikan bahwa OBI (diukur secara level-1) jarang melampaui 0.40—faktanya, >85% observasi berada di bawah 0.10. Akibatnya, dari 590 sinyal Golden Cross yang terpicu, 100% diblokir mentah-mentah sebelum dinilai oleh modul perhitungan rasio Sharpe/Kelly. Selain itu, strategi Sideways tercatat 0 kali aktif akibat ambang volatilitas kaku (0.085%) yang terlampau jauh dari kondisi pasar riil.

Real data proved that OBI (measured at level-1) rarely exceeds 0.40—in fact, >85% of observations sit below 0.10. Consequently, of the 590 triggered Golden Cross signals, 100% were bluntly blocked before being evaluated by the Sharpe/Kelly ratio module. Furthermore, the Sideways strategy triggered 0 times due to a rigid volatility threshold (0.085%) wildly detached from actual market conditions.

Proses verifikasi staleness menyingkap sebuah bug table mismatch krusial (crypto_klines vs btc_klines). Celah struktural ini menyebabkan instrumen inti QPS secara diam-diam bergantung sepenuhnya pada asupan data dari agen eksternal lainnya. Masalah telah diperbaiki pasca audit diagnostik, di mana mekanisme mandiri pembaruan data tiap 1-menit baru terverifikasi sempurna saat ini.

The staleness verification process revealed a crucial table mismatch bug (crypto_klines vs btc_klines). This structural flaw caused the core QPS instruments to silently depend entirely on data feeds from another external agent. The issue was rectified post diagnostic audit, with the independent 1-minute data update mechanism newly verified perfectly today.

Siklus perlindungan dari *Kelly Fractional* telah menyempurnakan bot. QPS dengan sadar menonaktifkan dirinya saat laju *Sharpe* memburuk tajam. Hal ini memotong drawdown raksasa yang umum dialami oleh sistem statis konvensional.

The protection cycle of the Kelly Fractional has perfected the bot. QPS consciously disables itself when the Sharpe rate deteriorates sharply. This cuts off giant drawdowns commonly experienced by conventional static systems.

Implementasi awal QPS Fase 5.0 mengandung kepincangan metodologi serius yang wajib diakui secara akademis. Ambang batas kritis (threshold) seperti penolakan OBI di 0.40, batas volatilitas batas di 0.085%, serta multiplier statis ATR (3.0x), ditegakkan berdasarkan asumsi kualitatif awal pengembang—bukan diekstraksi lewat analisis distribusi statistikal riil.

The initial implementation of QPS Phase 5.0 harbored serious methodological flaws that must be acknowledged academically. Critical thresholds such as the 0.40 OBI cutoff, the 0.085% volatility limit, and the static 3.0x ATR multiplier were established based on the developers' initial qualitative assumptions—not extracted through empirical statistical distribution analysis.

Lebih jauh, metrik OBI yang diterapkan sebetulnya hanyalah proksi likuiditas dangkal berbasis level-1 order book (sekadar best bid/ask), sangat jauh dari klaim agregasi institusional kedalaman order book menyeluruh. Tidak ada satupun tahapan walk-forward validation atau tes balik (backtest) formal komprehensif yang menjamin kelayakan parameter tersebut sebelum dilepas ke lingkungan eksekusi live.

Furthermore, the applied OBI metric was actually a shallow liquidity proxy based on the level-1 order book (merely best bid/ask), a far cry from claims of comprehensive institutional depth aggregation. Absolutely no walk-forward validation stages or rigorous formal backtesting processes vetted the viability of these parameters prior to their release into the live execution environment.

Penerapan indikator statis teknikal konvensional tidak lagi memadai di pasar mata uang kripto modern. Tingkat likuiditas buatan dan presisi manipulatif di pasar menghancurkan rasio keuntungan dari sistem yang bergerak terlambat (lagging).

The application of conventional technical static indicators is no longer sufficient in modern cryptocurrency markets. The level of artificial liquidity and manipulative precision in the market destroys the profit ratio of lagging systems.

Migrasi ke kerangka Quant Professional System (QPS) menyediakan kanvas rekayasa bernilai tinggi. Meski pendayagunaan proksi Order Book Imbalance (OBI) masih berada pada strata simplistis dan terhambat bias penetapan awal (Fase 5.0), integrasi pengereman proporsional rasional via kerangka manajemen Quarter Kelly Criterion telah memaku fondasi logis pertama bagi *bot* untuk matang. Penataan kalibrasi persentil empiris (Fase 5.1) yang kini dirintis diharapkan menutupi semua lubang metodologi tersebut.

Migration to the Quant Professional System (QPS) framework provides a highly valuable engineering canvas. Even though the utilization of the Order Book Imbalance (OBI) proxy remains in its simplistic infancy and hindered by initial designation biases (Phase 5.0), the integration of rational proportional braking via the Quarter Kelly Criterion management framework has nailed the first logical foundation for the bot to mature. The empirical percentile calibration restructuring (Phase 5.1) currently pioneered is expected to plug all aforementioned methodological gaps.

Pada Agustus 2026, sistem mengalami transformasi arsitektur fundamental: dari bot single-regime berbasis EMA Golden Cross menjadi sistem pemantauan multi-koin berbasis klasifikasi volatilitas dinamis. Fondasi Fase 5.2 adalah koneksi WebSocket All-Market Binance yang menerima data harga riil seluruh pasangan USDT secara bersamaan—termasuk dukungan pemrosesan pesan bertipe Binary Frame maupun Text Frame.

In August 2026, the system underwent a fundamental architectural transformation: from a single-regime bot based on EMA Golden Cross to a multi-coin monitoring system based on dynamic volatility classification. The foundation of Phase 5.2 is the Binance All-Market WebSocket connection that simultaneously receives real-time price data for all USDT pairs—including support for processing both Binary Frame and Text Frame message types.

Bot berlangganan satu saluran tunggal wss://stream.binance.com/stream?streams=!miniTicker@arr yang mengirimkan snapshot harga, volume 24 jam, perubahan harga harian, serta high/low harian untuk seluruh simbol aktif di Binance setiap detik. Data ini disimpan dalam struktur Arc<RwLock<HashMap<String, TickerData>>> yang dapat dibaca secara paralel oleh thread analisis tanpa blokir.

The bot subscribes to a single channel wss://stream.binance.com/stream?streams=!miniTicker@arr that delivers price snapshots, 24-hour volume, daily price changes, and daily high/low for all active Binance symbols every second. This data is stored in an Arc<RwLock<HashMap<String, TickerData>>> structure readable in parallel by analysis threads without blocking.

Sistem secara otomatis menghitung Standard Deviation 24 jam dari data klines per koin, kemudian mengelompokkannya ke empat tier volatilitas: EXTREME (≥3.0%), HYPER (1.5%–3.0%), HIGH (0.5%–1.5%), dan LOW (<0.5%). Parameter Trailing TP—meliputi trail trigger dan pullback threshold—dikalibrasi secara berbeda untuk masing-masing tier ini, sehingga strategi keluar posisi menyesuaikan diri secara adaptif dengan karakter pergerakan harga koin tersebut.

The system automatically calculates the 24-hour Standard Deviation from each coin's kline data, then categorizes them into four volatility tiers: EXTREME (≥3.0%), HYPER (1.5%–3.0%), HIGH (0.5%–1.5%), and LOW (<0.5%). Trailing TP parameters—including trail trigger and pullback threshold—are calibrated differently for each tier, so that the exit strategy adapts to the price movement characteristics of each coin.

Setiap siklus analisis (tiap menit), sistem menjalankan pipeline seleksi 4 tahap untuk memilih 100 koin yang paling potensial dari 400+ pasangan USDT aktif di Binance: (1) Filter Minimum: Harga >0 dan volume 24 jam >$100.000; (2) Skor Volatilitas: Peringkat berdasarkan StdDev 24 jam dari tertinggi ke terendah; (3) Skor Volume: Bobot volume digunakan sebagai faktor tiebreaker untuk koin dengan volatilitas setara; (4) Batas Maksimum: Dipilih Top-100 koin terbaik dari seluruh kategori volatilitas.

Each analysis cycle (per minute), the system runs a 4-stage selection pipeline to choose the 100 most promising coins from 400+ active USDT pairs on Binance: (1) Minimum Filter: Price >0 and 24h volume >$100,000; (2) Volatility Score: Ranked by 24h StdDev from highest to lowest; (3) Volume Score: Volume weight used as a tiebreaker for coins with equivalent volatility; (4) Maximum Limit: Top-100 best coins selected across all volatility categories.

Sejak diluncurkan pada 9 Agustus 2026, Bot A Fase 5.2 mencatatkan 25 transaksi SELL sukses dengan Win Rate 100% pada saldo awal simulasi $200. Seluruh transaksi keluar melalui mekanisme Trailing TP adaptif berbasis tier volatilitas—tanpa satu pun transaksi Stop Loss terpicu. Koin yang masuk ke dalam pemantauan aktif dirotasi secara dinamis berdasarkan kondisi pasar real-time, mencerminkan kelincahan fundamental dari arsitektur multi-koin ini dibanding versi bot sebelumnya yang terpaku pada koin tunggal.

Since its launch on August 9, 2026, Bot A Phase 5.2 recorded 25 successful SELL transactions with a 100% Win Rate on a $200 simulation starting balance. All exit transactions occurred through the adaptive Trailing TP mechanism based on volatility tier—with no Stop Loss transactions triggered. Coins entering active monitoring are dynamically rotated based on real-time market conditions, reflecting the fundamental agility of this multi-coin architecture compared to previous bot versions locked onto single coins.