Built my own backtester as a learning project - Open-sourced for feedback

I decided to build my own backtesting framework in C# (NovaTrader) primarily to learn how these systems work under the hood. Ended up implementing some features I couldn't find in existing tools. It is designed for portfolio based EOD strategies using market orders.

Features:

• Multi asset portfolios - use multiple assets at once
• Multi strategy portfolios - combine multiple strategies
• Hierarchical strategy nesting - Strategies within strategies, infinitely deep
• Online Portfolio Algorithms - FixedShare and ExponentiatedGradient for adaptive allocation without manual weight tuning.

Simple example:

Code:

public class RsiLowHigh(SecurityId sid, int period, double low, double high) : Strategy
{
    public int Lookback { get; init; } = 42;

    protected override void OnInitialize()
    {
         Benchmark = AddSecurity(sid);
    }

    protected override void OnUpdate()
    {
        var close = Benchmark!.GetHistory().Close.TakeLast(Lookback).ToArray();
        var rsi = TA.Rsi(close, period)[^1];
        OrderTargetPercent(Benchmark, rsi < low ? 1 : 0);
    }
}

//////////

// Adaptive example
var trend = new BuyAndHold("QQQ");
var meanRev = new RsiLowHigh("QQQ", 2, 25, 75);
var adaptive = new FixedShare(trend, meanRev) { Eta = 50, Alpha = 1e-4 };
AddSubStrategy(new ApplyPortfolio(fixedShare) { MinOrderValuePercent = 0.02m });

Built for EOD portfolio strategies using market orders only. C# framework, small and fast.

GitHub: https://github.com/trenki2/novatrader-core

Sharing this in case it's useful for anyone else. The hierarchical nesting feature seems unique - haven't seen it implemented elsewhere in a similar and easy way.

Looking for feedback from traders who might find these features helpful, or just curious if this fills any gaps in existing tools.

Thanks for taking a look.

 

nice project more though and detail than i have seen in a long time.

i was discussing this the other day with someone

• switching models based on profit factor deterioration
• dynamic allocation to highest performing pf model
• ideally switching off non performing models
• problem is idle model has to be live to track it
• goal is to minimize total portfolio drawdown

 

By nesting, you mean using objects ?

 

Yes, exactly. By “nesting,” I mean structuring strategies using the Decorator and Composite patterns. This allows a strategy to contain one or more sub-strategies, effectively forming a hierarchical tree of strategies. Each strategy can take the portfolio generated by its sub-strategies, apply additional rules or transformations, and produce a new portfolio.

This approach has several advantages:

1. Reusability: Individual strategies can be developed independently and reused in different combinations.
   
   
2. Flexibility: You can dynamically compose complex strategies by combining simpler ones without changing their internal logic.
   
   
3. Scalability: As your system grows, you can add new strategies without modifying existing ones, just by plugging them into the hierarchy.
   
   
4. Separation of Concerns: Each strategy focuses on a specific aspect of portfolio management, making the overall system easier to understand and maintain.
   

In practice, you might have a top-level strategy that applies overall portfolio constraints, while its sub-strategies handle individual sectors, risk adjustments, or optimization techniques. The final portfolio is the result of processing all these layers together.

 

Just a note on the requirements section on GitHub, you don't need to be on Windows or any specific IDE.
Since you are not using any GUI library and everything is console-based you can use it with any other environment. I am on Linux with Rider and can run it.

I can see that the Strategies project uses a UI, but that is just your choice to run it.

 

Ask @fan27 how that worked out for him.