Video summary
Работающие стратегии в алготрейдинге. Вся правда о торговле роботами. Павел Зайцев
Main summary
Key takeaways
Presenter / source context
- Interview with Pavel Zaitsev on algorithmic trading (“trading robots”), covering:
- how strategies are built, tested, portfolio-constructed, and risk-managed
- infrastructure choices and the need for retraining
Key ideas & recommendations
Building and evolving algo strategies
- Robots are not “set-and-forget.” They require active management:
- monitor which algorithms perform
- remove underperformers
- adjust exposure up/down as conditions change
- Retraining is essential because markets shift:
- retrain parameters on a weekly or monthly cadence
- Update “control parameters” using recent test results:
- examples include moving-average periods
- Donchian channel lookback ranges / min-max windows
- Simple strategies often outperform complex ones in practice:
- complex indicator/filter stacking is harder to generalize
- complex designs can overfit
- takeaway: lower complexity tends to be more robust
Trading approach across markets (progression)
Recommended progression:
- Start with futures (implied focus on a Moscow mini-futures segment first)
- Move to stocks on the Moscow Exchange
- Expand to St. Petersburg Exchange instruments (including fund/ETF-like products)
- Eventually consider cryptocurrency, but with cautions:
- high commissions
- opaque / less clear regulation
Methodology / step-by-step framework
- Stage 1: Idea generation & broad testing
- “Brainstorming approach”: test many trading ideas, keep the best performers.
- Stage 2: Build a “combat set”
- assemble a set of candidate algorithms
- filter out those with poor dynamics.
- Stage 3: Portfolio construction (robots as portfolio assets)
- use portfolio theory (Markowitz):
- choose a set of robots to balance return with minimizing drawdown
- select based on individual performance, then optimize the set together to reduce portfolio drawdown
- use portfolio theory (Markowitz):
- Stage 4: Deployment & risk-controlled scaling
- after the portfolio is running, increase trade volumes systematically as a percentage of deposit
- if performance deteriorates or “chaos” appears (e.g., dispersion across markets/robots), repeat:
- remove weak robots
- add robots with improved backtest/live test results
- Ongoing maintenance: Retraining
- retrain settings weekly or monthly using newly received data
Quantitative / numeric details & risk framing
Returns / risk ranges (high-level)
- Broad expectation range for algo trading:
- starting around a risk-free level of roughly 10–8% (subtitle text appears garbled; intent suggests a “roughly 10–8%” versus risk-free reference)
- up to about 30–40% per year on capital as an upper “typical” range (not guaranteed)
- Risk relationship emphasized:
- higher return generally implies higher risk
- no immunity from drawdowns (reference to “classic July ones”)
Retraining cadence
- Once per week or once per month
Performance metrics mentioned
- Drawdown
- Profit factor (explicitly named as a criterion)
- mentions profitability/return, profit dynamics, and filtering based on “poor dynamics”
Instruments / venues mentioned
Crypto
- Cryptocurrency (no specific coins/tickers named)
Stocks / exchanges / funds (Russia + index products)
- Moscow Exchange stocks
- St. Petersburg Exchange instruments
- S&P 500 index fund (as an example)
- Yandex (as an example underlying within the fund context)
- references to “mining companies” fund/slot
- mentions “funds/clubs” (subtitle wording is unclear)
Futures (Russia)
- Mini-futures on silver (explicit)
- references to the Moscow futures market
- mentions MMVB mini-futures on silver (subtitle includes “MMVB,” likely intended as part of MICEX/history)
High-frequency / microstructure
- mentions using:
- tick data from the order book (“lobby”)
- very short 7-second candles as an example test setup
Tickers
- No clearly legible specific stock/futures tickers from the subtitle text.
Strategy types and named setups
He repeatedly emphasizes a trend-following / breakout bias, with limited counter-trend experimentation.
Breakout / trend strategies (main)
- Breakouts:
- breakout of a moving-average “tunnel”
- breakout of the Donchian channel
- Pullback / rebound strategies within channels:
- “pullback candle” concept
- trading from channel boundaries (including rejection at boundary levels)
- Counter-trend / mean reversion (niche)
- he notes he is “slowly getting closer” to counter-trend strategies
- but they require specific instruments and tend to be more niche
Simplicity examples
- a very simple 3-candle directional concept (plus a “rollback” candle) described as surprisingly stable
- caution on indicator complexity:
- moving-average concepts (including “crossing moving averages”) may underperform simpler pattern/breakout concepts like MA tunnels
Infrastructure, risk controls, and cautions
Operational risks where robots fail
- internet/broker/order execution failures
- missing quotes / connection loss
- market gaps at open (“morning gap”) opening in the wrong direction
- exchange suspensions / widening limits and “bars” (large % moves) where orders may be rejected:
- in such cases, the robot may require manual intervention
- e.g., manual order placement when limits change or orders are not accepted
Execution / latency optimization (HFT)
For advanced execution, he notes options such as:
- direct connectivity to the exchange
- potentially:
- an own rented computer near exchange servers
- direct line / minimal latency
- typical stack notes:
- often Linux
- C# commonly mentioned
- some use low-level C for speed
Constraints and caveats he highlights for HFT:
- requires deep programming knowledge
- OS/system limitations can restrict “minute-scale” algorithm behaviors
- high transaction costs and expensive infrastructure
- claims only sufficiently capitalized players with adequate turnover negotiate better broker/exchange commission terms
Disclosures / disclaimers
- No explicit “not financial advice” disclaimer appears in the provided subtitle text.
Presenters / sources
- Pavel Zaitsev — interview guest; primary source of the content
- Interviewer — unnamed (only described via an introductory phrase, e.g., “Hello friends…”)