Automated Trading - FAQs

[IAS_LLC]

How is it possible,if the logic works with the dataset,side by side?

You must test with out of sample data, I.e train your algoritm with data from march and test it with data from june and july. If it falls apart, you are "overtrained", or curve fit.

 

[Brahmastra]

You must test with out of sample data, I.e train your algoritm with data from march and test it with data from june and july. If it falls apart, you are "overtrained", or curve fit.

But it still uses data series no matter how you twist and turn it.

 

[IAS_LLC]

But it still uses data series no matter how you twist and turn it.

Sure, but whats the problem with that? I cant think of anything in life that we do that isnt based on the "data series". Everything we do is based on past experiences and knowledge. Trading is no different

 

[eusdaiki]

But it still uses data series no matter how you twist and turn it.

All trading systems work around the time series of market data, and they seek to make probabilistic forecasts on the future data using present and past data from 1 or more sources.

this is true for any trading system, whether it is manually traded or automated...

 

[eusdaiki]

How is it possible,if the logic works with the dataset,side by side?

whether you are working on a simple system based on technical indicators, or a complex machine learning bot, the system must always have the ability to generalize to out of sample data. Failure to generalize to new data indicates that the system has over-fitted (it adapted to the particularities of the data sample not to exploitable patterns).
testing on out of sample data is standard procedure when developing algos that mine knowledge from the data, regardless of application. trading algos are not the exception. One of the reasons why walk fwd tests are considered a vital part in the preparation of an algo trading system is that the walk fwd test provides you with out of sample data and by its nature is immune to several biases that are hard to avoid on the backtest.

 

[Brahmastra]

I haven't seen such a system yet.

You haven`t seen them b/c all the retail stuff is bondaged with data series.When you use data series you curve fit from the begining.

i guess it is possible to create a logic that works with only 1 dataset

And how do you discard the data series?

 

[eusdaiki]

You haven`t seen them b/c all the retail stuff is bondaged with data series.When you use data series you curve fit from the begining.

And how do you discard the data series?

can you please explain what it is you're talking about?
I don't understand the concept of a trading system without market data, regardless of the system being automated or manually traded, it needs data to function. Whether that data is SEC filings, order book, trades, auction imbalances, or tweeter news... the system needs an input.

 

[Brahmastra]

can you please explain what it is you're talking about?
I don't understand the concept of a trading system without market data, regardless of the system being automated or manually traded, it needs data to function. Whether that data is SEC filings, order book, trades, auction imbalances, or tweeter news... the system needs an input.

I`m talking about any time series.Market data is the whole different story.

Order book will suffice.Does the order book need time series?If i want to look,for e.g.,for 264785836556973 block size,do i need data series?

 

[IAS_LLC]

I`m talking about any time series.Market data is the whole different story.

Order book will suffice.Does the order book need time series?If i want to look,for e.g.,for 264785836556973 block size,do i need data series?

You do need time series. The market is a dynamic system, d/dt . If d/dt = 0, we have no market. You can produce signals based on a single snapshot, but those signals are based on past knowledge. The exception where the past doesnt matter is arbitrage.

 

[eusdaiki]

Let's get our definitions straight. So that we all speak the same language.

MW dictionary defines a times series as:
: a set of data collected sequentially usually at fixed intervals of time.

Oxford dictionary provides this definition:
A series of values of a quantity obtained at successive times, often with equal intervals between them


All market data is a time series. Whether we are talking about a tradelist, orderbook, financial statements... it all becomes available as time moves fwd and can be sorted by timestamps. (it is obtained at successive times).

taking this definition into account, please explain what you mean when you speak of using no time series?


http://www.oxforddictionaries.com/definition/english/time-series
http://www.merriam-webster.com/dictionary/time series