Atchula and Pareek's exit comparison needs a fill audit before a trader can use it. Their experiment is clean, but its price series cannot be traded.
The bitcoin system trades Renko breakouts. Each Renko brick records a price move of a specified size, letting displacement rather than elapsed time drive the strategy. The script builds its own bricks, sized at ATR(14) times 1.0, instead of using TradingView's native Renko chart. A trend filter permits a direction; an exit rule determines when to leave. The authors wrote "Renko Trend Breakout Strategy 3.6.2" in Pine Script v5 and tested it in Strategy Tester, TradingView's backtesting engine. It takes long and short positions of one unit, starts with USD 100,000 and charges 0.01% commission per transaction. Its source is BITSTAMP:BTCUSD on one-minute Heikin Ashi bars, covering roughly 1 October 2025 to 2 October 2026.
Crossing six filters with two exits produces twelve runs. The filters are Renko higher-high/higher-low structure, EMA200, an EMA50/200 crossover, MACD(12,26,9), Supertrend(10, 3.0) and ADX(14) above 20. With the "base" exit, reverse logic makes an opposing signal the main route to changing position direction. The authors caution that the base still has risk management. The alternative adds a 400-point take-profit and a 100-point stop, plus a trailing stop that activates at 100 points and follows by 100.
The results change sign across the board. Every base run earns money: profits range from USD 38,442.20 for the crossover (PF 1.081) to USD 354,575.82 for ADX (PF 1.324). Every fixed-exit run loses, ranging from USD -223,595.20 for the crossover to USD -400,658.74 for Renko HH-HL. Their profit factors fall between 0.694 and 0.759. This consistency across filters is the authors' principal finding, and they qualify it carefully. The abstract warns readers against treating the results as evidence of live-trading profitability. In the discussion, the four proposed mechanisms remain hypotheses.
Enough parameters, missing trade rules
The paper gives a generous parameter list. It supplies the brick rule, numeric settings for five of the six filters, the four exit distances, position sizing, pyramiding of 1 and commission. The omissions concern how trades are generated. Several bricks can form within a source bar "subject to a safety cap", whose value we did not find. The HH-HL filter draws on "the most recent internal Renko bricks" without specifying how many. EMA200, the crossover and MACD explicitly use the internal Renko close; we did not find the input series for Supertrend or ADX stated.
The breakout entry never appears as an explicit rule. We also did not find units for "points" or expectancy. Given one-unit sizing on BTCUSD, we interpret both as dollars.
The code is proprietary and unreleased, according to the authors. A rebuild therefore tests a reconstruction of their concept. It cannot check the paper's twelve figures trade by trade against the original script.
Why do more winning trades still lose money?
Adding fixed exits compresses expectancy into much the same losing range across filters. The base produces 10.03 to 20.01 per trade; fixed-exit expectancy ranges from -11.10 to -12.67. Every filter trades more often under the fixed exit. The crossover rises from 2,540 to 18,902 trades, and ADX from 17,718 to 30,938. Four of the six filters also win more often, including the crossover, whose win rate increases from 29.57% base to 43.61% fixed. The exceptions are Renko HH-HL (37.18% to 34.25%) and ADX (38.18% to 34.42%).
The bracket arithmetic raises a question. The fixed-exit crossover wins 43.61% of its trades. If every exit occurred at the 400-point target or the 100-point stop, the calculation would be about 0.4361 x 400 minus 0.5639 x 100. That gives roughly +118 points per trade before costs. The paper reports -11.83. Many trades must therefore leave at other prices.
The trailing stop is a candidate: after activation at +100, it stands at entry. An opposing signal is another, because the paper describes the explicit configuration as adding the target, stop and trail to the base logic. Those additions leave the reverse logic in place. When a one-minute bar touches both the activation level and the stop, the simulator must infer their sequence.
We suspect that inference accounts for much of the result.
Commission could explain the other part of the gap. The tables cannot resolve that question. Among its metrics, the paper lists maximum drawdown and commission "where available in the exported reports". It also says the high-frequency runs "accumulated substantial simulated commissions". Neither results table contained a commission total or drawdown figure that we could find.
Losing USD 400,658 from a USD 100,000 account
The Renko HH-HL fixed-exit run records -400.66% on USD 100,000. Its fixed one-unit position continues trading after the account has been exhausted, four times over. The simulation imposes no margin call or liquidation. All six fixed-exit runs lose more than twice their starting capital: the crossover returns -223.60%, and Renko HH-HL reaches -400.66%. These figures are bookkeeping totals from a simulator that disregards capital. They need to be read on those terms.
The underlying prices deserve equal scrutiny. TradingView's own warning, cited by the paper, says Heikin Ashi strategy tests can produce unrealistic results. The authors also acknowledge that they do not claim the P&L "could be reproduced through live BTCUSD execution". We did not find a statement that the test forced fills onto standard OHLC prices. According to the TradingView note cited in the paper, non-standard charts use synthetic prices. This test takes simulated fills from that chart. Orders process on bar close, and recalculation after each simulated fill is enabled.
The authors present the backtest as a computational experiment on their specified chart representation. Within those limits, the abstract attributes a substantial difference in simulated behaviour to exit architecture. The fill model weakens that interpretation because the simulator handles the exit types differently.
Reverse-logic exits execute at a bar close. Targets, stops and trailing stops execute along an inferred path within the bar, using synthetic Heikin Ashi highs and lows. Changing that path would, in our expectation, affect fixed-exit runs more than base runs: only the fixed exits execute inside the bar. The sign flip may tell us as much about the broker emulator as about the strategy.
Our rebuild uses different fills
We are building and testing our own version now. Any result we publish belongs to our strategy and should be read accordingly. With the original script unavailable, we implement a Renko breakout independently using the disclosed parameters. Heikin Ashi bars supply signals only. Fills occur at the subsequent actual minute-bar price, a deliberate departure from the reported TradingView setup.
Minute bars leave the sequence unresolved when multiple bricks, trailing activation and competing exits occur inside the same bar. We therefore update signals only on completed bars and settle ambiguous fills against ourselves. Our history does not cover the full October 2025 to October 2026 window used in the paper. We test an overlapping stretch of BTC data instead, and we have not matched the Bitstamp venue exactly. We will also examine the target, stop and trail separately, retaining a holdout the parameters have never seen. The authors themselves identify out-of-sample periods and execution on actual prices as future work.
One result would change my reading. If the base exit remains profitable while the 400/100/100/100 bracket remains unprofitable on actual prices with conservative fills, an independent implementation would support the paper's interpretation of the exits. This time, the agreement would rest on actual prices.