The Hard Truth

The “95% of traders lose money” figure has long become a meme, but real data backs it up. Let’s look at why algorithmic trading – especially for retail participants – remains an extremely challenging endeavor.

MOEX Statistics

According to Moscow Exchange data and analysis from Habr:

• 76% of active traders on MOEX are unprofitable over a year
• Among those using algorithmic trading, about ~70% are unprofitable – slightly better, but not radically so
• Average retail algo trader loss: -12% annually (after commissions)
• Only 3-5% consistently profit over a 3+ year horizon

Hidden Costs That Kill Strategies

1. Exchange and Broker Commissions

Typical retail commissions on MOEX:

Exchange commission (stock market):
- Maker: 0.01% of trade volume
- Taker: 0.015% of trade volume

Broker commission:
- From 0.03% to 0.06% (depends on broker and plan)

Total round-trip (open + close):
- Minimum: 0.08% of volume
- Typical: 0.12-0.15% of volume

With 10 trades per day and an average position size of 100,000 rubles:

10 trades × 0.12% × 100,000 = 1,200 RUB/day
× 250 trading days = 300,000 RUB/year

That is 300,000 rubles per year in commissions alone. With a 1,000,000 ruble deposit, that is 30% annually that you need to earn just to break even.

2. Slippage

Slippage is the difference between the price your strategy “wanted” to enter at and the price at which the order actually filled:

• On liquid instruments (Sberbank, Gazprom): 0.01-0.05%
• On less liquid ones: 0.1-0.5%
• During news events: 1-5%+

3. Market Impact

If your order is significant relative to the order book, you move the price against yourself. This is rare for retail traders on liquid instruments, but a serious issue on thinly traded securities.

Why Backtests Lie

Look-ahead Bias

The most common mistake: using data that was not yet available at the time the decision was made. Examples:

• Using the day’s closing price to make a decision on that same day
• Using adjusted data that was changed retroactively

Survivorship Bias

Backtesting on S&P 500 stocks only accounts for companies that survived. Companies that went bankrupt or were acquired are not included in the test sample, creating the illusion of higher returns.

Overfitting

The most insidious enemy:

The more parameters in a strategy,
the better it performs on historical data
and the worse it performs in live markets.

If your strategy has 10+ parameters and shows 200% annual returns on backtest, it is most likely overfitted.

Regime Change

Markets change. A strategy that worked in 2020-2023 may completely stop working in 2024-2026. Examples:

• Volatility strategies designed before COVID broke down during the pandemic
• Momentum strategies tuned for a bull market lose in sideways markets
• Arbitrage strategies “close up” as more people copy them

The Real Cost of Algo Trading

Beyond trading costs:

Expense Item	Annual Cost
Server (VPS/colocation)	30,000 - 300,000 RUB
Data (historical + realtime)	10,000 - 100,000 RUB
Software (platform, tools)	0 - 50,000 RUB
Your own time	priceless

What to Do If You Still Want to Try

1. Start small – with a deposit you can afford to lose
2. Account for ALL costs in backtests – commissions, slippage, latency
3. Test on out-of-sample data – split history into training and testing sets
4. Limit parameter count – the simpler the strategy, the better
5. Use walk-forward analysis – regularly review parameters
6. Start with paper trading – test the strategy in real time without money
7. Diversify – don’t bet everything on a single strategy

Algo trading is not a “money button.” It is serious engineering and analytical work that demands discipline, capital, and honesty with yourself.