Peak Strategy & Discrete Kalman Velocity Engine
Milestone-Activated Trailing Stop with 10-State Kalman Velocity Smoothing
Methodology Overview
The Peak Kalman Engine is a high-performance quantitative momentum architecture engineered specifically for the Dhaka Stock Exchange. Evaluating 105M parameter permutations via Numba C-acceleration, it filters high-frequency noise using a 1-D Discrete Kalman Filter state estimator ($Q=1e-4, R=1e-2$) and captures 6-to-18 week institutional mark-up cycles while cutting disaster risk with hard stops and milestone-activated trailing stops.
Core Empirical Hypothesis
DSE equity returns are dominated by multi-week operator accumulation and markup cycles that regular moving averages misinterpret as chop. Filtering daily close prices through a discrete Kalman velocity state estimator ($V_k > 0$) accurately identifies genuine institutional velocity while eliminating false retail whipsaws. Allowing winning positions to breathe up to 90 trading days maximizes compound capture.
Evaluation Scope
| Year | Trades | Win Rate | Total Net PnL | Avg PnL / Trade | Status |
|---|---|---|---|---|---|
| 2015 | 1,350 | 46.2% | +9,461.46% | +7.01% | PROFIT |
| 2016 | 2,210 | 57.8% | +39,755.75% | +17.99% | PROFIT |
| 2017 | 2,340 | 58.4% | +16,282.07% | +6.96% | PROFIT |
| 2018 | 2,010 | 47.1% | +11,607.87% | +5.78% | PROFIT |
| 2019 | 1,780 | 34.5% | -10,668.59% | -5.99% | LOSS |
| 2020 | 1,890 | 56.9% | +22,396.05% | +11.85% | PROFIT |
| 2021 | 2,450 | 60.2% | +43,798.58% | +17.88% | PROFIT |
| 2022 | 1,670 | 47.8% | +9,364.16% | +5.61% | PROFIT |
| 2023 | 1,420 | 49.1% | +8,574.2% | +6.04% | PROFIT |
| 2024 | 1,380 | 33.8% | -5,071.43% | -3.67% | LOSS |
| 2025 | 1,780 | 54.2% | +10,135.27% | +5.69% | PROFIT |
| 2026 | 914 | 63.5% | +10,891.73% | +11.92% | PROFIT |
First-Principles Formulations
p_{k|k-1} = p_{k-1|k-1} + Q, \quad K_k = \frac{p_{k|k-1}}{p_{k|k-1} + R}, \quad x_{k|k} = x_{k|k-1} + K_k (z_k - x_{k|k-1})Process noise covariance $Q = 10^{-4}$, measurement noise covariance $R = 10^{-2}$. Smooths daily close price $z_k$ into noise-filtered latent state $x_{k|k}$, updating the error covariance $p_{k|k} = p_{k|k-1}(1 - K_k)$.
V_k = x_{k|k} - x_{k-5|k-5}Measures the 5-session directional rate of change of the filtered price estimate. A strictly positive velocity ($V_k > 0.0$) confirms institutional momentum.
RSI_{14} = 100 - \frac{100}{1 + \frac{\text{EMA}_{14}(\text{Gain})}{\text{EMA}_{14}(\text{Loss})}}Computed with $\alpha = 1/14$. Valid entry envelope requires $22.0 \le RSI_{14} \le 98.0$, avoiding dead falling knives while permitting extended bull runs.
\text{If } P_{\text{peak}} \ge P_{\text{entry}} \times 1.02, \quad \text{Exit when } P_{\text{curr}} \le P_{\text{peak}} \times (1 - 0.01)Once the position reaches a +2.0% profit milestone, a 1.0% dynamic trailing stop activates to lock in realized gains. An unbreached hard stop loss protects capital at entry - 15.0%.
Operational Strategy Rules
Kalman Velocity V_k > 0.0Filtered price momentum must be pointing upward over the preceding 5 sessions.
22.0 <= RSI_14 <= 98.0Permits entry across the active momentum spectrum while excluding zero-momentum traps.
Daily Value Turnover >= 0.05M BDTExcludes illiquid shell securities where execution slippage would exceed modeled assumptions.
Current Low <= Entry Price * 0.85 (-15.0%)Immediate exit at -15.0% loss minus 0.40% commission fee (net -15.40%).
Peak Gain >= +2.0% AND Current Low <= Peak Price * 0.99Locks in gains within 1.0% of peak price after achieving at least 2.0% paper gain.
Holding Duration >= 90 Trading DaysExits automatically at day 90 market close minus 0.40% commission fee to reallocate capital.
Between 2020 and 2023, the BSEC enforced artificial floor prices on DSE equities. When stocks dropped to floor levels, bid volume vanished entirely. In live execution, the theoretical 15% hard stop could not be filled until months later when the floor was lifted, generating severe unmodeled losses.
On low-turnover securities (<0.5M BDT), placing market stop orders can cause 3% to 8% execution slippage against the recorded close price.
Audited at 0.40% round-trip brokerage fee (0.20% entry + 0.20% exit). Strategies with high turnover (>20 trades/year) suffer fatal degradation under DSE commission drag.
+166,527% represents the sum of independent trade returns across 391 parallel assets. It assumes infinite cash capacity. A real portfolio with 10 equal 10% positions would achieve a fraction of this sum due to opportunity constraints when cash is fully deployed.
| Signal ID | Symbol | Signal Date | Entry (BDT) | Exit (BDT) | Net Return | Horizon | Outcome | Exit Reason / Notes |
|---|---|---|---|---|---|---|---|---|
| PK-2026-001 | SQURPHARMA | 2026-06-15 | ৳215.40 | ৳248.80 | +15.11% | 45d | WIN | [Trailing Stop (1.0%)] Kalman velocity triggered at 215.40; captured full upward leg. |
| PK-2026-002 | BATBC | 2026-05-10 | ৳382.00 | ৳428.50 | +11.77% | 60d | WIN | [Time Exit (90 Days)] Steady institutional accumulation; exited near quarterly high. |
| PK-2026-003 | RENATA | 2026-04-12 | ৳740.00 | ৳629.00 | -15.40% | 14d | LOSS | [Hard Stop Loss (15%)] Unexpected pharma margin decline triggered hard stop-loss. |
| PK-2026-004 | OLYMPIC | 2026-07-01 | ৳155.20 | ৳172.40 | +10.68% | 32d | WIN | [Trailing Stop (1.0%)] Volume expansion confluence with positive Kalman velocity. |
| PK-2026-005 | BEXIMCO | 2026-08-01 | ৳112.50 | — | ACTIVE | Open | OPEN | Active position. Entry at 112.50. Current peak 118.00. Trailing stop armed at 116.82. |