Algos outperform mostly due to the amount of trades they do, not their genius Only Jack Hershey was able to do 1000 trades a day profitably.
If you think algo trading is just applying indicators and backtesting, you've been misled. I wish it was that easy.
In the HFT space, this holds true. Outside of that, there is a multitude of clever designs that can trade as little as a few trades per day, depending on conditions.
There's not many people I don't respect, for what it's worth. I could write a book about proper algo design, but I'll try to answer your question concisely by listing what I feel are critical elements/requirements. Catalysts Volatility Volume Data mining Predictive modeling Signal processing Executable/Programmable Trade management Few degrees of freedom as possible Quantitative performance monitoring Slippage accuracy Fill accuracy That's just the tip of the iceberg. Predictive modeling, execution, fill/slippage accuracy were the most difficult during my research phase. Also, pay no attention to all the psychology/risk management edge bullshit. Without a good predictive model as a starting point, you'll be up shit creek as a retail trader. I learned this the hard way.
That's because it became much more competitive out there. Some 20 years ago, institutions, professional traders and middlemen made (more or less) reliable income from order flow, spreads, commissions. Nowadays, they have to compete against each other.
Once again, always dozens of master of the universe, claiming to have achieved GRAIL, and never a SINGLE specific data point . Always generalized bullshit any moron could come up with.. digitalnomad's post a perfect example no one is fooled