<a href="http://www.datatime.eu/public/gbot/Sun 29 Aug 2010_port4001_Cli5/GBotReport_2010-08-29_port4001_Cli5.htm"> Chart update </a> This week have been doing a lot of (nocturnal) work on code. Hopefully, next week we should be able to test different methods to overlay folios. For now i have let run the clone idle (testing). We made about 10K with a max exposition of 31 contracts. <img src="http://www.datatime.eu/public/gbot/Res_Sun29.png"> I am planning to try the following methods. Let me know if you have more ideas to test. 1. Start a folio normally and its "clone". Then activate the autotrading for the clones discretionarily, instrument by instrument, when one instrument shows a significant drawdown and is reversing. 2. Start a folio normally and its clone. Then activate the autotrading for the clones <b>automatically</b> and <b>constraining</b> an opposite position. (It could start trading at the same time as the other folio, or we could switch "autotrading" one clone instrument at a time: perhaps it's better to do this second way, so that the result can be compared with the previous method.) Ideally, if we had sufficient capital to keep alive multiple folios (4-5) (instead of stupidly increasing the initial packet size) we would be practically behaving similarly to MMs, working at all price levels, also providing liquidity to mkt. (Clearly we could also start multiple robot instances.) We would be gaining from all price moves, buying low and selling high. Tom
Exactly Hook, that's why we overlay, because we do not want to <b>increase the size of the initial packet</b>: it's certainly better to "spread" the risk on multiple overlaying folios with (out of sync) "opposite" positions. This way we reduce the risk (hedging). [The variance of results decreases: profit might also become relatively "slower", i expect] Tom
What a great project! You wrote: I do not understand why. So let's take an example. Let say your portfolio contains 2 securities: A (last quote USD50, realized volatility 10%) B (last quote USD25, realized volatility 30%) Assume the realized correlation between A and B is 0.75. G-Bot bought 100 A. So as to hedge the position, how many B will it sell?
Hi betcashrun, thanks and great nickname! volatility is a friend here, provided that it does not mean **extremely** large price moves. In fact, in this case the bot would "invest" more on the moving instrument. What we would like better is a global investment "balanced" among instruments, without exaggerated peaks. 1. So if there are instruments which tend to "track together", we can let just one go, like a little price "scout", while the correlated ones are "delayed" (in terms of entries). (Anyway the "scalp normalization" helps "breaking" pretty effectively much of correlation.) 2. The other powerful conceptual device i am proposing here (and want to refine) is <b>"folio superimposition"</b>, so that one istrument will be actually hedged by a copy (a "clone") of itself. The reason why i expect an outcome variance reduction is because the algorithm (profitability) is "invariant with respect to time shift", so we could start a "clone" in such a way it would (mostly) take opposite entries. This way, we would have an instrument and its "copy" both following a (possibly) profitable strategy, but with one "covering" the "half cycle" of the other one. Pretty much like if you had 2 traders playing a profitable strategy on the same account, but one of the trader will tend, "out of sync", to be short when the other one long. And if we overlay more folios and more robots (= more "traders"), we will be pretty much covering all the price range possibly making money from every moves: massive speculation and market making. With small packets size we will go also mostly "undetected" in the market ("stealth" mode). Finding the <b>best folio (auto) overlaying rules</b> is the next challenge! I like to talk here of "Investment" instead of "drawdown" because here no money is ever "irrecoverably lost". What i mean by that is that, with any strategy incorporating stop or stop/reverse at trade level, any time you "reverse", the corresponding amount of money is <b>** GONE FOREVER **</b> beyond recovery. It's a permanent loss. Here since we are overlaying a pair of trenders, there is <b>no money "permanently lost" </b> (unless of course one gets liquidated for shortage of capitals ;-). May be matter of time: but the investment is all there and eventually will turn into profit. "Drawdown" is actually a slightly misleading word in this context. To me it's actually more like an "investment". Tom
Thank you Tom. My point is that in case of a risk-adjusted porfolio, the expected risk on one instrument should be the same whatever the volatility of the instruments: the more volatile, the smaller the size.
Yes, i see your point. The question is actually more articulate. I think that the concept of volatility is simply too "broad" just to reduce the investment on an instrument if it shows a higher volatility figure. Volatility can take many forms, and the same figure can be attached to endless different price curves. For instance, we have seen that in the previous runs we have been making nice money on CL, which certainly wasn't the less volatile in the folio. So, would we really invest relatively less on it, maybe to overload ZB ? (I doubt). The unpleasant occurrences are not necessarily related with instrument volatility. Take May 6. So it might actually be that investing in a manner inversely proportional to "volatility" may not turn out to have always the expected consequences. Tom
LOL betcashrun! I am happy to know you have influential friends. I really need to expand my social relations ) Well, it does not seem to me that Prof. Sharpe is exactly saying "invest more on less volatile instruments". If you refer to his ratio, in the Sharpe Ratio, apart the return std at the denominator, you also have the average profit at numerator. This means that if to an higher return variance corresponds a much <b>higher average profit</b>, even the Sharpe Ratio might be higher. So, if you tell him, you may actually be supporting my point. ;-)) Besides, with all due respect to Prof. Sharpe, who has earned a Nobel prize, i take the chance to make a few consideration about the ratio itself, in algo trading context. In a general context of economic theory the ratio is a fine metric. But let's focus on algorithmical trading. Many quants try to accomplish the old saying "let your profit run". But what does that mean ? Letting the profits run may cause an increase (a "good" one) of return variance. But the Sharpe ratio places any variation at the denominator ("good" and "bad" variance), and this does not make it at all an ideal metric to evaluate trading algorithms. Another important consideration is that we really don't give much about return variance, as we are more interested in worst case scenarios, for <b>strategy calibration purposes</b>. I am often disappointed when talking with hedge fund managers one of the first things they ask me about is the Sharpe ratio of my algorithms. Though, I know it's an industry standard and people need a common language. (But i feel like have spent years thinking about possibly better and original ways, and then everything is reduced to the same old common places.) Tom
I agree with you: if you do not perform risk-adjusted performance analysis, worst case scenario analysis (i.e. value at risk) may prevent you from blowing up.