Moscow Institute of Physics and Technology (MIPT) has presented a system for testing cryptocurrency trading strategies that is built not on idealized price history, but on simulating conditions close to real trading. The development accounts for fees, slippage, liquidity, funding costs for perpetual contracts, and other factors due to which a profitable strategy on paper may turn out to be weak after launch.
Development Card
- Developer: Moscow Institute of Physics and Technology (MIPT).
- System premiere date: July 30, 2026.
- Industry: Financial services, investments, and auditing.
- Technologies: Data Mining.
The system belongs to data mining tools and is intended for stricter evaluation of cryptocurrency strategies. Its task is to show how an algorithm would behave not in laboratory conditions, but in an environment where a trade is executed with costs, delays, and market constraints.
What is Backtesting and Why is it Needed
Backtesting is the testing of a trading strategy on historical data. Simply put, the algorithm essentially runs through a past market segment again: it receives prices, opens and closes trades according to set rules, and then shows what the result could have been.
Such testing is needed to preliminarily evaluate the strategy's logic: whether it made a profit, how much the capital drawdown was, how often losing periods occurred, and whether the idea withstands different market phases. For a trading robot, backtesting is especially important: it helps identify weaknesses before launching with real money.
How to Backtest a Strategy or Robot
Testing is usually built according to a clear algorithm:
- Prepare historical data: prices, volumes, fees, and other market parameters.
- Configure the strategy rules: entry conditions, exit conditions, position size, and risk limits.
- Factor in real trading conditions: fees, slippage, liquidity, and funding costs for perpetual contracts.
- Run the test so that the algorithm does not use future data.
- Analyze the results through reports, charts, and trade tables.
For backtesting, platforms like MetaTrader, TradingView, Amibroker, and similar are often used. If a proprietary testing system is needed, it is usually written in Python, C++, or R.
Optimizing a trading robot boils down to enumerating and comparing parameters: for example, indicator period, exit rules, or position size. But choosing only the option with maximum profitability is dangerous — it's important to see how stable the result is across different market segments.
There are also additional verification methods. Forward testing runs the strategy on new data after configuration to test it outside the historical segment on which parameters were selected. Visual testing shows trades directly on the chart step by step: this makes it easier to see where the algorithm enters, exits, and makes mistakes.
Metrics, Factors, and Common Backtesting Errors
The backtest result is assessed not only by profit. Typically, several indicators are looked at simultaneously:
- Profitability: Shows how much the strategy earned over the selected period.
- Maximum Drawdown: Shows how much the capital decreased from peak to trough.
- Sharpe Ratio: Helps assess risk-adjusted return.
- Volatility: Shows how sharply the strategy's results changed.
- Profit/Risk Ratio: Helps understand if potential profit is worth potential losses.
The success of the testing is most strongly influenced by data quality, accounting for fees, liquidity, slippage, realism of simulation, length of the testing period, and risk of overfitting.
The most common errors are using future data, overfitting to a lucky historical segment, ignoring trading costs, and too short a testing period. It is precisely due to such errors that a strategy may look strong on a chart but quickly lose effectiveness after launch.
What Exactly MIPT Scientists Tested
The focus of the research was momentum strategies — one of the popular approaches in algorithmic trading. Their idea is simple: if an asset's price is already moving in a certain direction, it may continue that movement for some time. Such approaches have long been applied in traditional financial markets, but in cryptocurrencies, they are often tested in an overly simplified manner.
The crypto market features different types of strategies:
- Momentum: Betting on the continuation of an already started price movement.
- Arbitrage: Searching for price differences between platforms or instruments.
- Scalping: Frequent short-term trades with a small profit target.
- Trend strategies: Trading in the direction of the main market movement.
- Counter-trend strategies: Attempting to profit from a pullback against the current movement.
- Market Making: Placing buy and sell orders to profit from the spread.
The main problem with such tests is detachment from reality. The quote on a historical chart shows the price but does not reveal all trade conditions: what the liquidity was, how much execution cost, what the slippage turned out to be, and how market cycles changed. Because of this, a strategy may look strong on past data but quickly lose expected profitability after launch.
The team from the "Blockchain" department at MIPT decided to test how robust momentum strategies are under a more realistic approach. To do this, the work was divided into several practical steps:
- Creating a testing system.
- Collecting hourly spot market data.
- Collecting eight years of hourly perpetual futures data.
How the Testing System Works
The development step-by-step simulates the work of a trading strategy as if it were launched in real time. At each step, the system checks several conditions:
- The algorithm receives only the market data that could have been available at the moment of decision-making.
- The trade result is calculated considering fees.
- Slippage is added to the calculation.
- Available liquidity is taken into account.
- For perpetual contracts, the funding cost is accounted for.
After calculating one time segment, the algorithm moves to the next and repeats the procedure on updated data. This approach helps avoid a common error — using information from the future. As a result, the test shows not an abstract return on history, but a more plausible picture of the strategy's behavior in trading.
The system reproduces the strategy's operation step by step: it uses only data available at that moment, accounts for real trading costs, and then moves to the next period. This method allows assessing how the strategy could have behaved in actual market conditions, — explained Maxim Egorov, a graduate of the "Blockchain" department at MIPT.
For a market where traders often test ideas in MetaTrader, write their own models in Python, or migrate high-load algorithms to C++, not only the logic of entering and exiting a position is important. The quality of testing rests on three things:
- Accounting for costs that arise during trade execution.
- Accounting for execution constraints, including liquidity and slippage.
- Comparing the final assessment with how the strategy could have performed in real trading.
Results: Profitability Turned Out Not to Be the Main Advantage
The obtained results were additionally checked using statistical methods. This was needed to separate a sustainable effect from a random coincidence of parameters with a favorable market segment. Such control showed: many strategies that look good under standard testing noticeably lose effectiveness after accounting for real trading conditions.
The main conclusion was not that momentum strategies yield maximum profit. A more important advantage was their ability to limit losses during periods of sharp market declines. Under realistic testing, capital protection during crashes came to the forefront.
132 variants of the strategy were run through the system. The research showed that one cannot rely only on the single variant that performed best on historical data. After adjustment for random enumeration, its result was no longer statistically different from luck. However, the momentum effect itself did not disappear.
If the strategy is evaluated not by the single most successful scenario but on average and a portfolio of several variants is formed, it retains stable profitability. At the same time, its key property is reducing drawdown during crisis periods. According to MIPT calculations, the maximum drawdown was reduced by more than 2 times: from 67% to 29%.
Why Such Testing is Needed for the Crypto Market
The authors of the development believe that the system can become a basis for evaluating and creating new investment products in the cryptocurrency market. Its value lies in the fact that a product can be preliminarily tested under conditions close to real trading to understand how capable it is of protecting capital during strong declines.
Such a system can form the basis for evaluating new investment products for the cryptocurrency market. It allows one to see in advance how a strategy passes through periods of strong decline and how controlled the risk remains. Against the backdrop of the formation of a regulatory framework for digital financial assets in Russia, such tools may become especially in demand, — noted Maxim Egorov.
The next step for MIPT scientists is to test other classes of trading strategies. The research is also planned to be extended to new types of crypto assets to assess how realistic testing changes the understanding of risk and return in different market segments.
end-content
