How fast can LEAN run?

You will never be able to compete with out-of-the-box algorithmic trading software. Low-latency solutions are always custom, written specifically for a certain goal. If you have the kind of capital needed for the infrastructure to run a low-latency setup (co-located and all), you should also invest in some low-latency engineers who can build a dedicated server that responds sub-millisecond, and write the accompanying code. LEAN was not built for that purpose. Instead of joining the "race to zero", LEAN is all about being smarter. Backtesting has always been one of the greatest frustrations in algorithmic trading due to high data costs, long computation times and last but not least: the time it takes to program a backtest and successfully deploy your own backtesting framework (with statistical analysis and all). QC takes all of that trouble away so the users can simply focus on what's important: building the best strategy. I hope this clears things up :)

HI JP, VERY IMPRESSED with Lean, but I'm wondering what is the lowest number it runs at, so I know which strategy is right for Lean, Thanks.

Backtesting speed depends on your CPU, but people have seen 500k ticks per second with dense data. For live trading its roughly 1ms to get from the live data queue handler to the algorithm event. You could place an order and it might reach the brokerage API in another 1ms.

Good description, JPB. If you'd ever want to compete with HFT, you'd need to shell out billions to have your computer sit next to the exchange room on NYSE. Auto, what do you mean by "lowest number it runs at"? If you mean the expected granularity of the data's timeframe, it can theoretically be per-second. However I'd hesitate to put forward a strategy that trades at that speed. The smallest granularity I've personally programmed towards is per-minute, with best results on a per-hour or -day basis.

@Stephen - LEAN natively supports tick data streams as well, in both backtesting and live modes. The important thing (in my opinion) is if you're going to try and do an HFT strategy is to properly model expected delays in the fill models.

That's Great, guys faster enough for my strategy.

@JB Many of the HFTs are now using Field Programmable Gate Arrays (FPLA) on a board in a slot on the Exchange's computer, so the latency is nanoseconds. So there is no server and no processor, the algorithm is reduced to pure high speed logic. Idiot Savant but really really fast.

Pretty soon they'll have dedicated psychics on both ends that can enter trades with zero latency. :-P

Nice

@Stephen Distance arbitrage using quantum entanglement for quotes and trades.

http://www.zerohedge.com/news/2015-02-21/holy-grail-trading-crosses-twilight-zone-hft-firm-virtu-has-lost-money-once-1485-tra What does the technical architecture look like software and hardware? The Ex Head Trader now works for a different HFT company and they have a career page discussing what hardware and software they use. http://gtsx.com/ What do you think their structure is ? What's a individual small scale version of this do you think? Not trying to trade like HFT but speed is important.