Konsta

328 reputation

bio	website
	location
	age
visits	member for	1 year, 1 month
	seen	May 3 at 21:37
stats	profile views	24

	328 reputation	bio	website		visits	member for	1 year, 1 month
16 badges		location			seen	May 3 at 21:37

summary answers questions tags badges favorites bounties reputation activity

23 Comments

suggestions reviews revisions comments badges posts accepts all

Oct 20	comment	Alternative liquidity measures If you explain what Amhiud is, I might give you another one.
Oct 19	comment	.NET statistical packages recommendation It's not .NET or C(++/#), but you could use python or call it from these languages. There you could use pandas. If you can write C, you'll learn Python in an hour. Knowing Matlab would be a plus.
Oct 17	comment	Order and position management in (semi-)automated trading system Simple database: save sent orders, save broker/exchange response, i.e. logging. Positions are deduced from responses. History and currently active orders/positions should be separated (flag as active). Maybe I overcomplicated by thinking about design patterns.
Oct 14	comment	Resources for performance statistics of trading systems It's just monthly data, but maybe state of trend following is of interest to you. There are monthly posts at least for a couple of years. see automated-trading-system.com/…
Oct 13	comment	Resources for performance statistics of trading systems With 'successful trading systems' I see 2 difficulties: 1) 'successful' is debatetable and depends on your definition (is a year of non-negative returns enough, although ultimately you lost 75% in 3 years?) 2) I believe truely successful strategies are not disclosed and those that are, might suffer from survivorship bias. Maybe looking at (hedge) fund performances would be a viable alternative, although survivorship bias might apply as well.
Oct 10	comment	Analyzing tick data 1, 2, 3 & 4 are options, not steps. Either 1, 2, 3 or 4. ad 2) treat time as some variable tightly related to the original time series, maybe forecasting both in order to know where price goes and when it goes there. ad 3) find some small time increment such that all oservations of the original time series roughly fit on some time of your new equidistant time series. ad 4) summarize your data maybe per 500 microseconds and create for example open/high/low/close information for each 500 microsecond batch
Oct 9	comment	Switching from Matlab to Python for Quant Trading and Research While I strongly dislike R's syntax and Python gives this pragmatic feeling of "I can get things done [quickly]", Python really lacks R's statistics tools. Python has a mid-size basis of statistics tools, but you need to call R for pretty basic stuff like seasonality/trend-decomposition, not to speak of more advanced statistics, but it partly compares with Matlab, I think. I believe something like the Rmetrics package isn't given anywhere outside of commercial tools yet, so that is definately something Python is lacking currently.
Oct 9	comment	Backtesting VaR model violation independence Although you seem to refer to Value at Risk (VaR), I wasn't sure in the first half of your question if you maybe refer to Vector AutoRegressive (VAR) models.
Sep 24	comment	Trade Count Time Series I am undecided which answer to accept. Since I believe that IB offers a lot of asset classes and exchanges and the count is supposedly then available for all history, I opt for this one.
Sep 23	comment	Trade Count Time Series Ok, so IB provides this data. Nice. This is not an option if you do not trade via IB, but for me and, I believe, many others this is ok.
Sep 23	comment	Trade Count Time Series Yes, I stumbled upon this link yesterday and figured that simply loading tick data might give a hint - how to interpret ticks? It is not the BBO I get, but the prices of trades that happened, right?
Aug 18	comment	How to hedge against lack of volatility What is "the 30 delta" here? Is it delta of 0.3?
May 19	comment	How does one go from measure P to Q(risk-neutral) when modeling an asset paying dividends? S under Q is supposed to return r (risk-neutrality) in total. Since it continuously yields delta, a drift of r under Q would yield r+delta. Thus risk-neutral drift is to be corrected by -delta.
May 18	comment	How does one go from measure P to Q(risk-neutral) when modeling an asset paying dividends? Yes, you are right. It should've been mu=mu-r, i.e. theta=(mu-r)/sigma with mu*=0 with your gBm assumption. This would've yielded (r-delta) under X.
May 18	comment	How does one go from measure P to Q(risk-neutral) when modeling an asset paying dividends? ad) "I assume you are referring to dS/S = mu dt + sig dz, but why does this stop being gBm if mu = 0?": In "The index is described as "following a geometric Brownian motion", which to me says that the there is no other drift going on" you seemed to deduce that mu=0 from the fact that the index is supposed to follow a gBm. I wanted to make clear to you, that although the index is supposed to follow a gBm, this does not imply that mu=0.
May 2	comment	Good reference on sample autocorrelation? Your claim that lim_{n->inf} sqrt(n) r_hat is supposed to be an element of the normal distribution does not make sense mathematically. A distribution is not a set. If you mean its support, that is the real line, an element of which your limit surely is. But there is little value in that claim. Besides, regarding autocorrelation basically autoregressive time series and fractional Brownian motion (and numerical approximation thereof) come to my mind. But I'm not sure wether either topic fits your background and time frame.
Apr 30	comment	Stock Price Behavior and GARCH miggety: "at each time step": Is that for each path and each time step, so maybe 100,000 GARCH recalculations per time step? Does the GARCH-result depend that much on 1 additional time step? I mean you add 1 data point each time step. @Owe Jessen miggety's GBM formula is the Euler-Maruyama discretisation of the SDE which yields an analytic solution like the one you are using, Owe. See GBM as well.
Apr 27	comment	Is there an optimal covariance one would want forecasts to have? While "generate a time series of return forecasts" might be what could be done, in standard portfolio optimisation you take only into account the expected return for the next period. So you are suggesting to save these forcasts for a couple of periods and look at the covariance of these series? After some more reading and thinking your model pretty much looks like a VAR (vector autoregressive) model.
Apr 23	comment	What mathematical characteristics are required from the asset price process in order to stay within the RNP framework? Practicality is a difficult quality. I think you do not need to over-engineer a model. It should fit your underlying, payoff and purpose enough. Maybe it must be calculated in 1 ms instead of being accurate to the 7th decimal point. And no, PDE methods do not always help you. Their approximations might suffer from the "curse of dimensionality". So they might be incomputable for complicated underlying and payoff.
Apr 22	comment	What mathematical characteristics are required from the asset price process in order to stay within the RNP framework? If you can find a risk-free portfolio that replicates the conjured payoff, return must equal the risk-free interest rate. Else, an arbitrage opportunity exists.