Hi Marcello, first I need to know if you mean leverage 1:5 or 1:1,2. And assume operation is a buy or sell so 150 is 75 roundturns? For France you would have to take into account FInancialTransactionTax (FTT) so cfd's might work for you. So with more info people could give you better advice.

Hi Sybren, thanks for your answer. I mean leverage 1:5. Operations consist in changing the market exposure of the system almost every day, The exposure for each day goes from -30 to +30 pieces (or any multiple of that).

My problem with cfd is that my italian platform doesn't cover that stock, but I could move to a french broker. The stock is Air Liquide.

what about interactive broker with multichart platform to run your strategy?

Giuseppe, I now I.B. but I don't really need an advanced platform. I need the lowest execution costs. So, I'm realizing perhaps I need an institutional partner (investment bank or hedge fund) rather than a good broker.

if I may ask, how did you learn how to analyze these patterns. is there a good book that would give me insight into fear and greed indicators?

I assume this is an intraday trading algo. Have you back tested on tick data yet? Your strategy should work across multiple instruments in a similar way, only requiring adjustment of the main parameters.

Hi, i would linke to know if this can work for indian markets also. Could you please share the names of some good books we can read for becoming a successful trader please?

1:5 with a real execution (not some retail shop taking the other side) indeed forces you to look for a more institutional solution. If IB is too expensive (starts at 5 bps) think you need to look for another solution. But with this small edge and leverage your mdoel should be pretty good and you can def not afford to trade against a MM. CFD is a good way to avoid FTT on Air Liquide.

I say 150 trades during a year, or for a backtest is not sufficient to provide a usable assumption the strategy also works in the future - especially if optimized with parameters. Strategy development using daily bars is not possible - you would base 50-70% on fundamental data.

If you for example develop a strategy in the 1 minute timeframe and test one month, which is 60*24*60 =28000 bars --> 100 years of daily bars. Then test this strategy on other months and you get often a very different result.

Further problems trading stocks or stock indices is, they have been bullish since 2008 - thus all long only strategies work, probably only the long only strategies works. Besides, stocks can only be traded 8 hours a day, there are 16 hours where you are out of the market.

I would like to discuss your strategy further . Thank you .

Jon, this is an overnight strategy, working on daily closing prices. Sure I've tested it on tick data, also changing the time window.

Working on multiple instruments is not so easy. It's not just a matter of parameters. In my experience, a family of patterns work just on a very limited number of assets, not necessary of the same type. I give you an example: fear & greed works on just one Forex pair, one stock index and a few stocks.

Taha, my trading systems at the time work on Forex Exchange and european and american indices and stocks (big companies). As long as you can acces those markets from India, they should work as well. I've no experience of the indian stock market itself, maybe there are opportunities as well with algo trading.

The best book I've ever found about trading is "Way of the Turtle: The Secret Methods that Turned Ordinary People into Legendary Traders" by Curtis Faith. I appreciated the fact that it focuses on the trader behavior rather than the trading systems.

My personal recommendation is creating the systems by yourself, keeping them simple and understandable.

Søren, my systems try to discover the computational dynamics of market prices, so they don't use fundamental data. The problem of sustainability over time is a big one, and it's strictly connected with the overfitting problem. I use a genetic algo for optimising the parameters, so, theoretically, the same system should be able to dadpt to different market conditions. The best method to verify this assumption is to repeat the entire cycle of training, testing and simulate operatios over time. The problem in doing so are the huge computational power and time needed.

Garette, the strategy is based on three trading systems. The first is a bundle of moving averages, each of which has a weight (trading quantity) and a direction (long or short). The second is a fear & greed indicator that buys or sells when prices are below or above a threshold. The third is a pattern recognizer that looks for simple differences in the closing prices in last n days.

The strategy combines those trading systems by means of a consensus mechanism, generating for each day a signal with quantity and direction of the position to take.

Conner, you are addressing a real big challenge in machine learning techniques: you can obtain good results without being able to explain them. We could think this doesn't matter, as long as the results are useful, and assume that the algorithm is catching some hidden dynamics in the phenomenon (e.g. a stock price series).

The only issue that we face in doing so is overfitting, and yes I agree about testing on multiple assets to try to reduce it.

I tink many have an unrealistic thinking and expectation of automated trading - you cant "just" start developing strategies - there is a long and steep learning curve using TRIAL and ERROR, time is THE key factor wether you will/may reach something usfull some years ahead. Minium 10.000´s of trial and error is needed to pass the learning curve, probably rather 100.000´s.