Video summary

A Estratégia de Médias Móveis 37x79 Explicada | A ARTE DO TRADE

Main summary

Key takeaways

Finance

Finance-focused summary (moving-average strategy + quantitative risk/process)

Instruments / tickers mentioned

  • Mini-index (Ibovespa Mini / “mini index”): primary instrument discussed (noted use of 1-minute bars and multiple timeframes).
  • No specific equity tickers, ETFs, bonds, commodities, or FX pairs were clearly identified in the subtitles. Mentions like “corn” appeared, but without actionable ticker/contract details.

Strategy framework: “MMK averages” (asymmetric moving-average crossovers)

The presenters describe an automated trading approach based on simple moving averages (SMAs) with:

  • different parameters for buy vs. sell, and
  • strict execution timing rules.

Core rules / methodology

  • Use simple moving averages (SMAs).
  • Apply asymmetry:
    • Use two SMAs for buying and two SMAs for selling.
    • Crossovers are side-specific:
      • When buy SMAs cross → buy
      • When sell SMAs cross → sell
      • Signals that would trigger the opposite side are ignored.
  • OHLC execution timing:
    • On crossover detection, the system waits and executes at the open of the next candle (using OHLC), not immediately on the crossover bar.
  • Explicit parameter set called out:
    • 37/79 (SMA periods), described as:
      • 37 = fast
      • 79 = slow (for the relevant side/setup)

Risk/Reward (explicit example)

  • Take profit (TP): 800 points
  • Stop loss (SL): 400 points
  • Implied 2:1 reward-to-risk (even though win rate is below 50%).

Performance metrics / numbers cited

Win rate and trade frequency

  • Success rate for the 37/79 example: 45.56%
  • Trade frequency:
    • Since 2025: 169 trades
    • Over roughly 270 trading sessions (low occurrence)
    • Another variant mentioned: 331 operations in one year, attributed to changes like fewer crossover checks when adjusting timeframe/average length.

Profit per trade (fee-aware)

  • Example: R$ 2,445 over 331 trades ⇒ R$ 7.39/trade
  • They emphasize ensuring a “cushion” against small costs (e.g., slippage/spread), aiming for average profit-per-trade around > ~7.

Additional strategy mention

  • A variant with SL 200 / TP 500 (trend context) and stated success rate ~38%.

Stop management / “training stop” (risk management concept)

They discuss a dynamic stop method:

  • Training stop:
    • Start with an initial stop loss (example: -400 points).
    • As price moves favorably, move the stop loss toward breakeven or into profit to protect gains.
  • Trend-based stop placement:
    • Stop placement based on the low of the last 5 bars minus an offset.
    • Mentions incorporating ATR (example phrasing: “according to ATR…”, with “five” referenced).

Explicit cautions

  • Overfitting is repeatedly highlighted:
    • Example rationale: changing a parameter slightly (e.g., 79 → 80) can flip outcomes from positive to negative, showing fragility.
  • Avoid “tinkering”:
    • Narrow time windows can produce a “perfect” setting that fails out-of-sample.

Overfitting control / validation approach (“time machine” backtest)

They propose a workflow to reduce optimization bias.

Step-by-step validation framework

  1. Start with 5 years of historical data (mini-index).
  2. Remove the last year (12 months):
    • Optimize/build the strategy on the first 4 years.
  3. Validate on the held-out last year using their normal validation tests.
  4. Apply a final pass/fail rule:
    • Test the last year with a yes/no criterion.
    • Example: reject even if results are only “one decimal place down.”

Time machine test (bias-free out-of-sample simulation)

  • Pretend you’re at the start of 2026, but only use data through early 2025.
  • Do not “touch” later data to avoid contamination.
  • After building, decide whether you’d put the strategy online today with a maximum live drawdown tolerance:
    • losses up to about R$200 were mentioned as unacceptable beyond that.

Important technical rule

  • The final “yes/no” test must happen only once. If parameters are changed after failure, the process corrupts the data.

Portfolio construction & correlation/delinking strategies

They emphasize that strategies should not be treated as isolated bets.

Correlation / decorrelation emphasis

  • They discuss combining trend-following and reversal approaches.
  • Core principle: prefer low correlation (decorrelation) so portfolio results are less dependent on a single market regime.
  • Example methods mentioned:
    • One presenter tests correlation across many candidate strategies and selects a set with low mutual correlation (including a correlation tool/program anecdote involving ChatGPT).
    • Another presenter constructs strategies expected to be opposites, then confirms low correlation via testing.

Operational advice

  • When signals conflict (buy vs sell), the system chooses behavior based on market context (trend vs sideways/regime).
  • One strategy mentioned can go ~3 months without trades, framed as a potential feature of the design: “nothing happening” can still be intentional.

Explicit risk management / Monte Carlo + position sizing

They outline portfolio-level sizing using backtests plus Monte Carlo testing.

Step-by-step risk framework

  • From backtests, identify the maximum loss over a multi-year horizon (stated: last 5 years).
  • Run a Monte Carlo test:
    • described as a probability test that changes operations over time without manipulating values
    • references a Gaussian approximation and standard deviation concept
  • Use Monte Carlo outputs to set investable amounts, allowing an emotional buffer.

Example portfolio (“Galaxy”)

  • Recommended investment: R$ 6,000
  • Historically most lost: R$ 300
  • “Max loss + two standard deviations”: R$ 4,000
  • Additional leeway logic:
    • clients might already be down about R$ 3,000; without buffer they might stop investing emotionally.

Additional constraints

  • Limit number of trades per day (stated: at most ~2, often ~1 preferred).
  • If crossovers whipsaw repeatedly, the strategy should not run indefinitely.

Recommendations & key takeaways

  • Use asymmetric moving-average setups (different SMA periods/logic for buy vs sell) rather than a single symmetric parameter set.
  • Execute at the next-candle open rather than immediately at the crossover tick.
  • Expect win rates < 50% when payoff structure is designed to compensate (example: 45.56% with TP 800 / SL 400).
  • Focus on net average profit per trade after costs:
    • aim around > ~7 profit/trade to withstand fees/slippage.
  • Avoid overfitting:
    • use parameter stability checks and strong out-of-sample validation (time machine).
  • Build a portfolio with low correlation across strategies (e.g., trend + reversal), not a single isolated bet.
  • Size risk using Monte Carlo + standard deviation framing, including psychological drawdown tolerance.

Disclaimers / disclosures

  • The subtitles include phrases like “this is no joke” and general educational discussion.
  • No explicit “not financial advice” disclaimer appeared in the provided subtitles.

Presenters / sources mentioned

  • Caio Scot
  • Márcio Killing (host)
  • Artur (Outlier Invest co-founder/trader)
  • Felipe (Outlier Invest co-founder/trader)
  • Defusco (referenced as discussing buying vs selling specialization)
  • Financial Market Wizards (book referenced)
  • Thomas Bulkowski (book/cyclopedia referenced)
  • Outliers (book referenced by title/name in the conversation)
  • Mentions of ChatGPT (used in a programming/correlation anecdote)

Original video