Robuxio Equities is launching: 1/ The key question: how to compound across different market regimes.

Many algorithmic traders optimize their strategies for backtest returns. The metric that actually determines whether your model survives in production is something else entirely.

If you actually tested running your strategy with and without stop losses, the result would probably surprise you.

AI cannot build robust models completely autonomously. Yet. But the rate of improvement in AI capabilities is exponential. There is no good reason why it shouldn't get there. What will happen then? That’s a question I’m spending a lot of time thinking about.

One of the interesting questions we are solving these days is standalone vs portfolio performance of a model. A strategy's standalone PnL is not its result in the portfolio. When dozens of models compete for the same slots (on the limited tradable crypto universe), the one ranked last waits and gets the worst positions. Priority logic matters. But every logic layer on top of all models is also a new way to break everything.

Every guide to execution in crypto talks about TWAP and VWAP. We solved the slippage problem in our portfolios a completely different way, and it improved diversification at the same time.

Most people think running an algorithmic crypto fund is mostly about trading strategy. The truth is that the majority of the work has almost nothing to do with trading.

Latest by the end of 2026, @robuxio_com will have AI integrated to all main process families: - software development - trading research - regular portfolio evaluation - business development - business operations - business support We are already pretty far, still some way to go. That's the only way to continue growing as efficiently as possible.

The most common pushback I get on crypto portfolios is how do you diversify when everything moves together? Here's the answer.

Be very careful with sharing your API keys with any 3rd party service. Many are being built in crypto to evaluate the trading results. Reverse engineering of single strategy approaches is very possible.

The problem of retail traders is that they think more about what other retail traders do. But the biggest issue is that all/most of the edges retail traders exploit are usually well known to institutions where the real money and liquidity is. Institutions are the ones who will cause the quicker edge decay. I do think that dumb retail group with AI tools without trading knowledge can start creating different market inefficiencies. Example: More people will start trading short-term breakouts on crypto which can cause steeper short-term inefficiencies which can be exploited with mean reversion models.

Lately, I'm thinking a lot about AI, the market, and trading edges. What will probably happen? My base scenario: What we can see with AI in general is that things are changing quickly. We also know that edges do appear and disappear. Of course, it depends on whether they are beta-driven or alpha-driven. Beta-driven edges tend to stay longer, maybe the factor is getting weaker over the long term. Alpha-driven edges are more the competition game and they are getting arbed away with the growing competition. I believe that in the age of AI edges will disappear much faster, especially simple alpha-driven ones. They'll be exploited quicker to their liquidity limits. And trading broad portfolios while working on new edges in a semi-automated way is the only way to survive going forward. The market is already pretty competitive, and it will be more competitive than ever. Who won't use AI heavily will have hard times.

Another good piece from @Jacknomics, our CGO. youtu.be/r5QLUptUyKw?is=vgLR…

Risk management in crypto means defending against four specific risks.

Travel hack: Save Asia for winter. Go to the Peloponnese in October/early November.

Research after AI development: In the past, the bottleneck was the number of hypotheses we could test. In the future, the bottleneck will be our ability to distinguish real edges from the results of massive data mining.

Most retail traders compare their live results to their original backtest. We compare them to something completely different.

Most retail algo traders turn off a strategy after one bad month. A real fund's process for actually retiring a model looks nothing like that.