Nick Radge – Daily Journal

[Nick Radge]

I have been asked to speak at a Quant conference in Singapore in early November. I had to highlight to the organizers that I only had a high school education – every other speaker has some type of PhD or multiple physics/math/statistics degrees.

For example, here is one speakers topic:

“…follow the market by solving a pair of coupled Hamilton–Jacobi–Bellman equations; we can construct sparse mean reverting baskets by solving semi-definite optimization problems with cardinality constraints and can optimally trade these baskets by solving stochastic control problems; we can identify statistical arbitrage opportunities by analyzing the volatility process of a stochastic asset at different frequencies; we can compute the optimal placements of market and limit orders by solving combined singular and impulse control problems which leads to novel and difficult to solve quasi-variational inequalities.”

:blush:

[JulianCohen]

Nick Radge wrote:

I have been asked to speak at a Quant conference in Singapore in early November. I had to highlight to the organizers that I only had a high school education – every other speaker has some type of PhD or multiple physics/math/statistics degrees.

For example, here is one speakers topic:

“…follow the market by solving a pair of coupled Hamilton–Jacobi–Bellman equations; we can construct sparse mean reverting baskets by solving semi-definite optimization problems with cardinality constraints and can optimally trade these baskets by solving stochastic control problems; we can identify statistical arbitrage opportunities by analyzing the volatility process of a stochastic asset at different frequencies; we can compute the optimal placements of market and limit orders by solving combined singular and impulse control problems which leads to novel and difficult to solve quasi-variational inequalities.”

WTF !

:blush:

[Anonymous]

you could talk about constructing sparse mean-reverting baskets on the mentor course!

[TrentRothall]

Nick Radge wrote:

I have been asked to speak at a Quant conference in Singapore in early November. I had to highlight to the organizers that I only had a high school education – every other speaker has some type of PhD or multiple physics/math/statistics degrees.

For example, here is one speakers topic:

“…follow the market by solving a pair of coupled Hamilton–Jacobi–Bellman equations; we can construct sparse mean reverting baskets by solving semi-definite optimization problems with cardinality constraints and can optimally trade these baskets by solving stochastic control problems; we can identify statistical arbitrage opportunities by analyzing the volatility process of a stochastic asset at different frequencies; we can compute the optimal placements of market and limit orders by solving combined singular and impulse control problems which leads to novel and difficult to solve quasi-variational inequalities.”

:blush:

That cracks me up

I often wonder about so called “Quants”… They all tlk in that sort of language and are clearly incredibly smart but does it translate into success? What sort of yearly returns are these guys generating? +100%?

You hear guys like Jerry Parker talk about simplicity is best etc etc but then these quants talk in language like that.. Would be interesting to know how they do…

[JulianCohen]

You might find a lot of measurebating going on

[LeeDanello]

Nick Radge wrote:

I have been asked to speak at a Quant conference in Singapore in early November. I had to highlight to the organizers that I only had a high school education – every other speaker has some type of PhD or multiple physics/math/statistics degrees.

For example, here is one speakers topic:

“…follow the market by solving a pair of coupled Hamilton–Jacobi–Bellman equations; we can construct sparse mean reverting baskets by solving semi-definite optimization problems with cardinality constraints and can optimally trade these baskets by solving stochastic control problems; we can identify statistical arbitrage opportunities by analyzing the volatility process of a stochastic asset at different frequencies; we can compute the optimal placements of market and limit orders by solving combined singular and impulse control problems which leads to novel and difficult to solve quasi-variational inequalities.”

Has he made any money trading??

Scientific concepts are just approximations using curve fitting techniques.

Your methods and approximations would probably be more robust as you’d have less variables and linkages that can break. I think KISS works best especially for the ones that can identify with their ignorance.
A lot of academics are in love with their qualifications.

[ScottMcNab]

Nick Radge wrote:

I have been asked to speak at a Quant conference in Singapore in early November. I had to highlight to the organizers that I only had a high school education – every other speaker has some type of PhD or multiple physics/math/statistics degrees.

For example, here is one speakers topic:

“…follow the market by solving a pair of coupled Hamilton–Jacobi–Bellman equations; we can construct sparse mean reverting baskets by solving semi-definite optimization problems with cardinality constraints and can optimally trade these baskets by solving stochastic control problems; we can identify statistical arbitrage opportunities by analyzing the volatility process of a stochastic asset at different frequencies; we can compute the optimal placements of market and limit orders by solving combined singular and impulse control problems which leads to novel and difficult to solve quasi-variational inequalities.”

Might be nice to have a point of difference…many speakers may be quants who are involved in trading as opposed to a trader who uses quantitative trading techniques ?

:blush:

[Nick Radge]

Sep:

ASX Swing: -0.04% (first losing month this year)
Growth Portfolio: I don’t measure monthly results but the system hit new equity highs late in the month.

US HFT: -0.38%
US MOC: -1.28% (been tweaking this during the month)
US Momo: +2.91%

[Nick Radge]

I have made some tweaks to my US MOC system and have started the live trading.

In prior testing my real time results were ALWAYS at the low end of the MCS, both on a daily basis and cumulative.

To be somewhat more aligned to the ‘average’ of the distribution I decided to really restrict the universe to the NASDAQ 100 from the S&P 500.

Results declined a little, but are still acceptable.

The second step was to decipher how many possible signals are available on any given day. Variability is from 5 through to 80, where 80 is reasonably rare yet 40 are more common.

As such I have adopted an adaptive position sizing method whereby I want to maximise my allocation to each trade whilst trying to keep the distribution of outcomes down.

I tested using 40 positions at 5% and am happy with the results (CAGR 23% maxDD 11%) but obviously there are periods where there are less signals so I’m not being overly efficient with the capital.

I then tested using 20 positions and 10% and happy with the results (CAGR 49% maxDD 23%).

So these are the two ‘book ends’. What I intend to do in real time is adjust the allocation between 5% and 10% depending on the number of position to a maximum of 40.

Example, if I have 8 signals, they all get 10%. If I have 25 signals they’ll get 8% and if I get 40 signals they’ll all get 5%.

[ScottMcNab]

I have just in the last few minutes finished entering last month of trades into STT and my live results are at about the level of the lower end of CAR with MC set to 20% trade skip….and all the “reduction” in performance has come from just a few few days since the start of Aug where possible trades far outweighed the number of possible positions. All other days agreed with backtest to the cent. Looks like I need to try and tighten universe or increase positions available too or accept the variability of those days where a huge number of possible entries are triggered….currently using S&P500.

[ScottMcNab]

Nick Radge wrote:

So these are the two ‘book ends’. What I intend to do in real time is adjust the allocation between 5% and 10% depending on the number of position to a maximum of 40.

Example, if I have 8 signals, they all get 10%. If I have 25 signals they’ll get 8% and if I get 40 signals they’ll all get 5%.

Will a new version of batchtrader do this Nick ?

eg current version with system of 20 postions at 10%
46 positions entered on market open at 10%
batchtrader immediately sells last 26 positions to close them to reduce to 20 positions at 10%

new version….again 46 positions entered on market open at 10%
batchtrader sells last 6 positions to close them and then sells half of all of the remaining 40 positions to end up with 40 positions at 5%

[SaidBitar]

Nick Radge wrote:

I have made some tweaks to my US MOC system and have started the live trading.

In prior testing my real time results were ALWAYS at the low end of the MCS, both on a daily basis and cumulative.

To be somewhat more aligned to the ‘average’ of the distribution I decided to really restrict the universe to the NASDAQ 100 from the S&P 500.

Results declined a little, but are still acceptable.

The second step was to decipher how many possible signals are available on any given day. Variability is from 5 through to 80, where 80 is reasonably rare yet 40 are more common.

As such I have adopted an adaptive position sizing method whereby I want to maximise my allocation to each trade whilst trying to keep the distribution of outcomes down.

I tested using 40 positions at 5% and am happy with the results (CAGR 23% maxDD 11%) but obviously there are periods where there are less signals so I’m not being overly efficient with the capital.

I then tested using 20 positions and 10% and happy with the results (CAGR 49% maxDD 23%).

So these are the two ‘book ends’. What I intend to do in real time is adjust the allocation between 5% and 10% depending on the number of position to a maximum of 40.

Example, if I have 8 signals, they all get 10%. If I have 25 signals they’ll get 8% and if I get 40 signals they’ll all get 5%.

I liked the Idea and wanted to try it, so i modified the codes to check the number of signals per day and to adjust the number of positions accordingly. then i ran 4000 MCs to compare with my previous MCs.
i was expecting to find the STDEV tighter but the result was almost the same at leas the stdev for CAR and DD.

maybe i need to check deeper

[Nick Radge]

Scott – no. You’re over complicating the issue. The position sizing is done before the orders are loaded into BatchTrader.

Last night the system only generated 8 orders again, so each was allocated 10%. Has nothing to do with BatchTrader

[Nick Radge]

Said – did you use the much smaller universe?

[SaidBitar]

No I tested on S&P 500
I will test it on smaller universe

[Author]

Posts