Trading An Algorithm w/No Experience Putting Trades Through

sprstpd · Dec 19, 2015

windwine said:
try quantconnect.com they are faster compared to quantopian and also free for backtest.

Plus they have tick data.

Yuliia Stoliarova · Dec 20, 2015

Dear Members of Elite Trader,

My name is Yuliia Stoliarova. Currently I’m studying “Financial Risk Management”. Before I leave Aix-Marseille University, I need few minutes of your attention to help me with my Master’s thesis questionnaire about impact of rumors on financial markets. Your response is the unique contribution to my work. I will really appreciate your time and effort.
http://www.surveygizmo.eu/s3/90007472/The-impact-of-rumors-on-the-financial-markets

Thank you very much for your cooperation.

If it is possible, I kindly ask you to spread this questionnaire among your professional connections who deal with financial markets.

tradingcomputer · Dec 28, 2015

syswizard said:
Exactly....
and these seemingly "minor" restrictions actually make a big difference in performance of algos.

To be successful in this game, everything has got to be "perfect":
1) a good, non-correlated mix of algos
2) streaming data in the correct bar type
3) excellent risk management
4) a good bet-sizing layover
5) low commission rates
6) robust and reliable programmable platform
7) solid analytics/performance measurement tools

Did I forget anything ?

Thanks for sharing the list above.
Would it be possible for you to recommend a good data feed vendor?

syswizard · Dec 30, 2015

tradingcomputer said:
Thanks for sharing the list above.
Would it be possible for you to recommend a good data feed vendor?

There is really nothing wrong with the Interactive Brokers API for data acquisition except getting price history is tough because they "throttle" it.

With IB, you have two options in real time:
1) Real Time Bars
2) Sampled Subscription

tradingcomputer · Dec 30, 2015

Thanks for your response, I am using IB now, coded a work around to manage throttling by adding time delay.

Thinking to also get IQFeed when I have my simple setup working. Well IB is almost free, while IQFeed will cost at least $130.

syswizard said:
There is really nothing wrong with the Interactive Brokers API for data acquisition except getting price history is tough because they "throttle" it.

With IB, you have two options in real time:
1) Real Time Bars
2) Sampled Subscription

syswizard · Dec 31, 2015

tradingcomputer said:
Thinking to also get IQFeed when I have my simple setup working. Well IB is almost free, while IQFeed will cost at least $130.

I always thought about the dangers of having signals and indicators working from one data feed while the orders were placed using a different feed.
Can you imagine the signal indicating to buy stop at 40.50 and the IB data indicates 40.80 and you get filled 30 cents from where you expected.

tradingcomputer · Dec 31, 2015

syswizard said:
I always thought about the dangers of having signals and indicators working from one data feed while the orders were placed using a different feed.
Can you imagine the signal indicating to buy stop at 40.50 and the IB data indicates 40.80 and you get filled 30 cents from where you expected.

Oh, haven't thought about this, thanks for bringing this up, it is really an important point.

Considering this point,

If I decide to use another data feed, such as DTN-IQFeed or DTN-Nanex, it will be only historical data for back-testing and strategy discovery purpose, no order execution.
If these other feeds involve in real time order execution, for example searching the right stock to buy, the final decision should be made based on IB data, which add complexity to order execution, need to weight whether the value added by the other data-feeds good enough to justify additional complexity.

Negative Theta · Jan 1, 2016

aucociscokid said:
Would members here recommend we - with no trading experience - embark on algorithmically trading our own strategies with millions of our own money? What are some of the pitfalls of trading an algorithm? It's not just ALL automatic, for instance, is it? The algorithm does everything?

Hi,
Latency seems to be quite a hot topic over here and a little overhyped in my opinion. If you don't do HFT, forget latency.

Beside all the classic cliches of trading, I would recommend to pay attention to a thing which you just pointed:
"It's not just ALL automatic, for instance, is it? The algorithm does everything?"

You are right, it is very tough to create 100% autonomous algorithm. Simply put it doesn't work like that. Too much effort.
The workaround? You need some sort of discretionary input into your algo. Why? Because algos are very bad at identifying certain aspects of the market. For instance: what is the mood/status of the security you are trading, what are the market expectations? Ofc you could get some informations by reverse the pricing of certain derivatives instruments which are connect with the instrument that you trade. The first one which comes into my mind is the implied volatility from the options market.
But, you as a human can spot very "easily" because you input into your brain tons of discretionary inputs based on the price that you are seeing and also very important, the fundamental informations that is the real driver behind of that security.
Ignoring the very small microscructure moves [which can be exploited by HFT], the market is not moving because of a funny algorithm, instead the move behind was generated by an idea digested by all the participants in the market.
Even in a 30 seconds TF, the returns will still "favour" a side of the market because it will be "sucked" by the idea/news flow that I was talking above.

syswizard · Jan 4, 2016

Negative Theta said:
Hi,
Latency seems to be quite a hot topic over here and a little overhyped in my opinion. If you don't do HFT, forget latency.

Once again, what is the definition of HFT ? Is it based on average P&L ? Based on Time-in-trade ? Trades per day ?

Negative Theta said:
You are right, it is very tough to create 100% autonomous algorithm. Simply put it doesn't work like that. Too much effort.
The workaround? You need some sort of discretionary input into your algo. Why? Because algos are very bad at identifying certain aspects of the market. For instance: what is the mood/status of the security you are trading, what are the market expectations? Ofc you could get some informations by reverse the pricing of certain derivatives instruments which are connect with the instrument that you trade. The first one which comes into my mind is the implied volatility from the options market.

Suddenly, you are back to mostly discretionary day trading...where only a very few talented traders succeed.
My feeling is that your best approach is a "portfolio" of algo trading apps.....and then perform a small amount of discretionary trading as well.

Monitoring algo and discretionary performance and then determining and applying optimal capital allocation is the most important task at hand.