# Tag Info

25

You could for example look at this research paper released by Deutsche Bank's Research group just yesterday which defines both high-frequency and ultra-high-frequency trading. In the paper it says Typically, a high frequency trader would not hold a position open for more than a few seconds. Empirical evidence reveals that the average U.S. stock is ...

22

Strictly speaking, data snooping is not the same as in-sample vs out-of-sample model selection and testing, but has to deal with sequential or multiple tests of hypothesis based on the same data set. To quote Halbert White: Data snooping occurs when a given set of data is used more than once for purposes of inference or model selection. When such ...

21

The Medallion Fund doesn't take outside investors. They returned the original investor money years ago. So: if it's a Ponzi scheme, then they've figured out how to profit by ripping themselves off. That's nice work if you can get it.

18

Building an effective backtest is not significantly different than building any other kind of predictive model. The goal is to have similar behavior out of sample as you have in sample. As such, there are methodologies developed in statistics and machine learning that can be useful: Understand the bias/variance tradeoff. This is covered in many places. ...

16

This is practically a textbook case begging for the Kelly criterion. In your specific example, the optimal trade size is $f^*A$, where $f^*$ maximizes the average rate of return $$\mathbb{E}[\log (X)]=0.5\log(1+0.3f)+0.5\log(1-0.23f).$$ Here $f$ is the fraction of the current capital to trade. A straightforward calculation yields that f^*=\frac{0.3-0.23}{...

14

Below, I see a lot of support and resistance. Here's the code: x <- cumsum(rnorm(1000)) plot(x, type="l", main="Support and Resistance") Edit (03/03/2011) ================================================ Gortaur, I put my answer here to avoid filling up the comment area. Your question 1) "......I was not asking for the "garbage" literature, I can ...

14

A survey by FinAlternatives in 2009 concluded that "86% believe that the term “high-frequency trading” referred strictly to holding periods of only one day or less." (Aldridge 2009): There are two problems with this survey for our present discussion: (1) the meaning of the term has been clarified significantly since that survey and (2) it surveyed a wide ...

14

My definition is not pretty, but it's practical: If you trade based on 5- or 10-minute bars, I call that high-frequency trading. If you trade based on tick-by-tick data, including bids and offers, I call that ultra-high frequency trading. (Trading 1-minute bars is somewhere in between. Trading more slowly than 10-minute bars is "day trading".) I make this ...

13

I. Re: # of trades... According to WK Selph (former quant turned blogger) @ WK's High Frequency Trading How To: To give some idea of the data volumes, the Nasdaq TotalView ITCH feed, which is every event in every instrument traded on the Nasdaq, can have data rates of 20+ gigabytes/day with spikes of 3 megabytes/second or more. The ...

12

There are plenty of market models -- capital asset pricing model (CAPM), conditional CAPM (CCAPM), intertemporal CAPM (ICAPM), and arbitrage pricing theory (APT). But any model, finance or otherwise, requires assumptions. Under these models the market may pay you to play your strategy, but in return you must accept risk. So with one of these models you could ...

11

There are other strategy types not covered by mean-reversion/trend following: arbitrage - keep correlated assets close in price (SPX index versus the 500 stocks contained in it, or Gold trading in London versus Gold trading in New York) market making - buy on bid, sell on ask, gain the spread liquidity rebate - some venus pay you for putting limit orders ...

11

There are a some information about Renaissance Technologies available in The Quants from Patterson. Basically, and it's also what I heard in general, they are using intensively algorithmic trading, and from what I understood there are using Information Theory (they worked with Shannon if I remember well). I'd say it'd be harsh to say it's the next Madoff ...

9

I have seen Hansen's SPA ('Superior Predictive Ability') test and stepwise variants used for this purpose. Hansen's test is a Studentized version of White's Reality Check. The stepwise variants allow one to accept or reject the null of no predictive ability on a subset of some tested strategies while maintaining a familywise error rate. In his book, '...

8

There is no official taxonomy of quant trading models. After all, "valuations" are inherently subjective, no matter how much math we put behind them. But there are some industry-standard terms that might be helpful. Inside the Black Box has the following break-down: Price Trend Reversal Fundamental Yield Growth Quality It's also possible to ...

7

HFT can be loosely defined as any strategy where your profitability is a function of latency.

7

This blog post points to a presentation about backtesting and data snooping: http://www.portfolioprobe.com/2010/11/05/backtesting-almost-wordless/ I think the only non-datasnooping method there is is to trade live. But the problem of data snooping can be reduced by seeing how significant the backtest result is compared to what would have happened if the ...

7

The output of your model will be a realization of your assumptions. Shane's given you a great answer. Besides doing out of sample testing (i.e., calibrating on period X then testing in period Y only using info available at the time of each trade), I would add that you should test it in sub-periods. If you have a big chunk of data, break it up and see how it ...

6

Your question's title suggests the market prices are mean reverting. I strongly suggest verifying that assumption via one of the usual tests, such as the Augmented Dickey-Fuller test (implemented in the tseries package of R by the adf.test function, and in other R packages, too). If the market is truly mean reverting, a possible strategy is Detrend the ...

6

I would conjecture that the reason a proof does not seem to exist, is that in a purely theoretical framework, such a model could exist for 'corner cases'. Under the assumption that the existence of a meta-model would modify the usage of the model / effect the market, something like Russell's paradox would seem to occur. Except in the case where a stable ...

6

The defining characteristic of "high-frequency" is not the number of trades, but instead it is the number of orders you place, and in particular how often you are changing those orders. The scratch rate (cancel/fill ratio) is often very high. For every 1,000 orders you place, you might get 5 fills. This is the single most defining criteria of whether someone ...

6

"There is no secret sauce!" - Inside the Black Box: The Simple Truth About Quantitative Trading, by Rishi K.Narang In this book, which is well worth reading to get a good conceptual overview of the different components of a quant trading system, the author tells about "one of the most successful" quant funds hiring only the best academic researchers and ...

5

Contrary to popular belief, there does exist some truly high quality academic literature on this topic. The most sophisticated and well executed paper in this regard is Lo, Mamaysky, and Wang (2000). They write: In this paper, we propose a systematic and automatic approach to technical pattern recognition using nonparametric kernel regression, and we ...

5

5

I don't think that it is a real applicable trading system but it is more general work concerning the connection between chaos and financial markets. A good starting point is this (relatively recent) article: http://deepeco.ucsd.edu/~george/publications/08_ecology_bankers.pdf You can find his publications here: http://sio.ucsd.edu/Profile/gsugihara#pubs

5

So those are cumulative pnl figures and you are interested in the percent changes in pnl from one data point to the next? Don't use log returns, simply generate the percent changes through r(t)/r(t-1)-1. 4.3922/5.2735-1 = -16.71% (in your example time series I made the assumption that the time series is in ascending order. Given your description of the ...

5

You can point out to your friend that, statistically speaking, having more observations reduces uncertainty in estimators. Mathematically, $SE_\bar{x}\ = \frac{s}{\sqrt{n}}$, showing that the standard error of a statistical estimator decreases with increased observations. This argument is concise and consistent with the Taleb quote. From wikipedia on ...

4

Clearly, it's much more difficult than for a white-box strategy. But you still have some information: What is the return profile? What is the average holding period? Does it go long/short? What assets are traded? Now you can choose a benchmark of an index that matches these criteria as closely as possible. If an appropriate benchmark doesn't exist, ...

4

The map is not the territory, any model is an abstraction and will never be complete, the only complete model of the market is the market itself, and so on. I agree that this leads directly into Gödel, Turing, the halting problem, and other basic computability concepts. Try this thought experiment: Imagine the market as a turing machine named M, reading ...

4

I think the key to fund performance is the use of own money, not borrowed. In this case, it is possible to implement strategies that ordinary hedge funds can not use due to risk management.

4

For me, I would calculate daily returns for such a series by backing out the daily PnL and dividing by some volatility number. lets define your cumsum as "c_pnl": daily_pnl = c_pnl - [0; c_pnl(1:length(c_pnl-1)] max_draw = max(cummax(c_pnl) - c_pnl) pct_returns = daily_pnl / max_draw # in terms of drawdown Don't you have capital already in the ...

Only top voted, non community-wiki answers of a minimum length are eligible