Any constructive feedback would be appreciated. Thanks. Quantitative Trading System: General Framework1) Trade Setup: Prepare the time series data by extracting the invariant. Estimate the invariant using probability distributions to find the highest probability trade setup. 2) Trade Trigger: Generate a signal (Computational method + Threshold level) Examples of computational methods: Correlation, Cointegration, Cointelation, Machine Learning algorithm, Neural Network 3) Money Management System: Combine with the signal to get a full trading algorithm Position sizing, stops, profit exit levels, adjustments.4) Testing: Backtest the system. Check for overfitting. Realtime Papertrading Test. Forwardtest with Monte Carlo generated data.

QuoteOriginally posted by: outrunIf its high frequency and orderbook based strategies then you also need to look at various scenario's of how your "backtest order fills" would interact with real market.Thank you. This system will not be high frequency.

Its a very generic recipe, so its a bit hard to comment. A bit like trying to describe how to build a car by saying it has 4 wheels, some controls and you sit in it and drive. However one thing sprang to mind. What you refer to are not probabilities. They are something else. And here in lies the key. In this system what is your edge?Something that is a real edge (i.e. not beta with some crazy dispersion characteristics masquerading as alpha).The ingredients you list are certainly required to implement an edge, but don't necessarily constitute it.So what is the phenomenon you have identified in which you take risk but that gives a more stable risk adjusted return than all other participants in the market?Of course I don't expect you to say this, but it is the real question.What is your capital? What is your horizon? What instruments? Does the cost of liquidity wipe out the supposed edge (i.e. neglected or misunderstood liquidity demand costs masquerading as alpha)?

QuoteOriginally posted by: neuroguyIts a very generic recipe, so its a bit hard to comment. A bit like trying to describe how to build a car by saying it has 4 wheels, some controls and you sit in it and drive. However one thing sprang to mind. What you refer to are not probabilities. They are something else. And here in lies the key. In this system what is your edge?Something that is a real edge (i.e. not beta with some crazy dispersion characteristics masquerading as alpha).The ingredients you list are certainly required to implement an edge, but don't necessarily constitute it.So what is the phenomenon you have identified in which you take risk but that gives a more stable risk adjusted return than all other participants in the market?Of course I don't expect you to say this, but it is the real question.What is your capital? What is your horizon? What instruments? Does the cost of liquidity wipe out the supposed edge (i.e. neglected or misunderstood liquidity demand costs masquerading as alpha)?Instruments: SPX and VIX Options. Equity Options with High IV and volume.Capital: EnoughCost of Liquidity: LowSupposed Edge: System Identifies and sells High IV Options. For low volatility environments the system sells OTM Spreads when there is a high probability of continued low volatility in the underlying for the near term.Risk Management: Draconian

You said you look for the high probability trade setup, by that do you mean that you are looking for mispricing?Not really sure what you mean by the 'invariant' here.Anyway, you say 'check for overfitting' but maybe you want to place a specific penalisation into step 1 to make sure that you are not overly confident in any estimations made by your model. Otherwise I would think that this strategy exposes you to sudden losses in the event of any chages to market behavior.If you are basing the strategy on two regimes, you might want to look at having an explicitly regime switching model if you have not done that already. What sharpe are you targetting?

QuoteOriginally posted by: neuroguyYou said you look for the high probability trade setup, by that do you mean that you are looking for mispricing?Not really sure what you mean by the 'invariant' here.>> Not mispricing, but unusually high IV that based on historical levels doesn't stay that high for too long. >> I mean time invariant. Transform time series data to mostly i.i.d data with little or no autocorrelation. Anyway, you say 'check for overfitting' but maybe you want to place a specific penalisation into step 1 to make sure that you are not overly confident in any estimations made by your model. Otherwise I would think that this strategy exposes you to sudden losses in the event of any chages to market behavior.>> I like the PBO method by Marcos. There is always a nonzero probability that I could be wrong for a specific trade. That is where my money management strategy limits losses. If you are basing the strategy on two regimes, you might want to look at having an explicitly regime switching model if you have not done that already. What sharpe are you targetting?>> Correct. I'm looking into the MSM Volatility model as a possible switching strategy. I'm not big on regular sharpe ratios because they don't adjust for fat tail risk. Perhaps the Probabilistic Sharpe Ratio (PSR) may help. I'm aiming for a PSR of 3.0, MAXDD at 6%

QuoteOriginally posted by: AnalyticalVegaQuoteOriginally posted by: neuroguyYou said you look for the high probability trade setup, by that do you mean that you are looking for mispricing?Not really sure what you mean by the 'invariant' here.>> Not mispricing, but unusually high IV that based on historical levels doesn't stay that high for too long. >> I mean time invariant. Transform time series data to mostly i.i.d data with little or no autocorrelation. Anyway, you say 'check for overfitting' but maybe you want to place a specific penalisation into step 1 to make sure that you are not overly confident in any estimations made by your model. Otherwise I would think that this strategy exposes you to sudden losses in the event of any chages to market behavior.>> I like the PBO method by Marcos. There is always a nonzero probability that I could be wrong for a specific trade. That is where my money management strategy limits losses. If you are basing the strategy on two regimes, you might want to look at having an explicitly regime switching model if you have not done that already. What sharpe are you targetting?>> Correct. I'm looking into the MSM Volatility model as a possible switching strategy. I'm not big on regular sharpe ratios because they don't adjust for fat tail risk. Perhaps the Probabilistic Sharpe Ratio (PSR) may help. I'm aiming for a PSR of 3.0, MAXDD at 6% Very interesting. Couple of questions then:1. That Sharpe ratio seems rather high (for a low frequency strategy) particularly if you attest to such rigourous money management. What makes you sure that you are not underestimating your model risk? Would it not be better to target a lower Sharpe with a less agressive implementation?2. So it sounds like your strategy is based on mean reversion of IV. This relies on you being able to implement trades within the 'tails' of your invariant. You say liquidity is high and costs are low. But is this true even within this targeted region of IV? Sure these trades might be profitable if you can catch them, but can that be done? What is slippage impact?

QuoteOriginally posted by: neuroguyQuoteOriginally posted by: AnalyticalVegaQuoteOriginally posted by: neuroguyYou said you look for the high probability trade setup, by that do you mean that you are looking for mispricing?Not really sure what you mean by the 'invariant' here.>> Not mispricing, but unusually high IV that based on historical levels doesn't stay that high for too long. >> I mean time invariant. Transform time series data to mostly i.i.d data with little or no autocorrelation. Anyway, you say 'check for overfitting' but maybe you want to place a specific penalisation into step 1 to make sure that you are not overly confident in any estimations made by your model. Otherwise I would think that this strategy exposes you to sudden losses in the event of any chages to market behavior.>> I like the PBO method by Marcos. There is always a nonzero probability that I could be wrong for a specific trade. That is where my money management strategy limits losses. If you are basing the strategy on two regimes, you might want to look at having an explicitly regime switching model if you have not done that already. What sharpe are you targetting?>> Correct. I'm looking into the MSM Volatility model as a possible switching strategy. I'm not big on regular sharpe ratios because they don't adjust for fat tail risk. Perhaps the Probabilistic Sharpe Ratio (PSR) may help. I'm aiming for a PSR of 3.0, MAXDD at 6% Very interesting. Couple of questions then:1. That Sharpe ratio seems rather high (for a low frequency strategy) particularly if you attest to such rigourous money management. What makes you sure that you are not underestimating your model risk? Would it not be better to target a lower Sharpe with a less agressive implementation?2. So it sounds like your strategy is based on mean reversion of IV. This relies on you being able to implement trades within the 'tails' of your invariant. You say liquidity is high and costs are low. But is this true even within this targeted region of IV? Sure these trades might be profitable if you can catch them, but can that be done? What is slippage impact?1) I'm never naked short options. That keeps the risk down. The rest is all position sizing and adjustment. 2) Rising IV in high volume option contracts means demand for the options is increasing. They are very liquid so you can catch them. Slippage is very minimal. The trick is to make sure there is enough volume. Another thing to check is if IV is really high and not be fooled by high IVRank based on the last year.

You are making pure noise with this thread, come on. Yes this is the setup of any technical trading system and so what?QuoteOriginally posted by: AnalyticalVegaAny constructive feedback would be appreciated. Thanks. Quantitative Trading System: General Framework1) Trade Setup: Prepare the time series data by extracting the invariant. Estimate the invariant using probability distributions to find the highest probability trade setup. 2) Trade Trigger: Generate a signal (Computational method + Threshold level) Examples of computational methods: Correlation, Cointegration, Cointelation, Machine Learning algorithm, Neural Network 3) Money Management System: Combine with the signal to get a full trading algorithm Position sizing, stops, profit exit levels, adjustments.4) Testing: Backtest the system. Check for overfitting. Realtime Papertrading Test. Forwardtest with Monte Carlo generated data.

QuoteOriginally posted by: mtsmYou are making pure noise with this thread, come on. Yes this is the setup of any technical trading system and so what?QuoteOriginally posted by: AnalyticalVegaAny constructive feedback would be appreciated. Thanks. Quantitative Trading System: General Framework1) Trade Setup: Prepare the time series data by extracting the invariant. Estimate the invariant using probability distributions to find the highest probability trade setup. 2) Trade Trigger: Generate a signal (Computational method + Threshold level) Examples of computational methods: Correlation, Cointegration, Cointelation, Machine Learning algorithm, Neural Network 3) Money Management System: Combine with the signal to get a full trading algorithm Position sizing, stops, profit exit levels, adjustments.4) Testing: Backtest the system. Check for overfitting. Realtime Papertrading Test. Forwardtest with Monte Carlo generated data.it is pure noise to those who do not understand, as it should be.

AnalyticalVega, Have you taken a look at genetic algorithms? They seem like a reasonable way to optimize your variables(Position sizing, stops, profit exit levels, etc).http://en.wikipedia.org/wiki/Glowworm_s ... timization

QuoteOriginally posted by: Bentam64AnalyticalVega, Have you taken a look at genetic algorithms? They seem like a reasonable way to optimize your variables(Position sizing, stops, profit exit levels, etc).http://en.wikipedia.org/wiki/Glowworm_s ... izationYou need to be very careful as most of these optimization algorithms are very prone to overfitting.