投稿時間:2022-01-09 06:11:54 RSSフィード2022-01-09 06:00 分まとめ(12件)

カテゴリー等	サイト名等	記事タイトル・トレンドワード等	リンクURL	頻出ワード・要約等/検索ボリューム	登録日
TECH	Engadget Japanese	2018年1月9日、au春モデルとして「Qua Phone QZ」などが発表されました：今日は何の日？	https://japanese.engadget.com/today-203015284.html	quaphoneqz	2022-01-08 20:30:15
海外TECH	MakeUseOf	The 9 Best Google Sheets Formatting Tips for Creating Professional-Looking Spreadsheets	https://www.makeuseof.com/google-sheets-formatting-tips-create-professional-spreadsheets/	google	2022-01-08 20:16:39
海外TECH	DEV Community	Trading Algos - 5 Key Metrics and How to Implement Them in Python	https://dev.to/blankly/trading-algos-5-key-metrics-and-how-to-implement-them-in-python-5gb8	Trading Algos Key Metrics and How to Implement Them in PythonMetrics surround us Whether you re building the next big thing and need to measure customer churn retention rates etc you will always need numbers to really back it up This is the same when we are building trading strategies We need easy and straightforward ways of analyzing whether or not our model is actually doing well Now before we go into understanding various portfolio metrics we have to first really understand the point of having good metrics And for that we have to understand what it means for a model to do well For some of you readers out there you probably already know what you want out of a strategy and what metrics you may want to do For that feel free to skip ahead to the various metrics and their implementations What is Doing Well Now when you think of a good investment strategy what do you think the most important thing is First and foremost at least for most people it is I want to make the most amount of money possible i e your return on investment which can come in many forms too as we will soon see But then this begs the question of a multitude of things when do I get my returns how do I get my returns do I care about my returns every year and arguably one of the MOST important questions what are my chances of NOT getting my returns i e my risk Ahh you see now the question gets much more nuanced as there are a lot more factors So I propose a different statement of doing well Doing well is not just about how much money you make but also about when you make it and how much risk you are taking on A model does well when it successfully produces high returns year over year while reducing the amount of risk for those returns That s awesome now how do we actually measure that How do we measure risk returns and all of these other factors Well that s why we have metrics Introducing Strategy MetricsNow for many people new to quant or even pros probably don t know all of these metrics and each of these metrics might be used for a variety of use cases Before we go into them I think it s super important to keep in mind that not one metric tells an entire story just like a person is not just governed by their name No one metric can tell you whether or not a stock is a good buy or not nor can one metric tell you if your model is good or bad So let s get into it We re going to go through useful metrics in this post If you want to see more metrics or don t want to implement all of these yourself you can take a look at the Blankly package that is one of the fastest and easiest to set up trading packages out there Also here are some great explanations on some of the metrics that we go over today I ve implemented a gist that houses all of the metrics you ll see below Compound Annual Growth Rate Starting Easy Compound Annual Growth Rate or CAGR for short is one of the most important metrics and the most traditional metrics out there What does the CAGR tell us Well it takes our cumulative return how much did we cumulatively make across X years of a backtest or a strategy and finds the percentage that makes it such that your cumultaive return was equal to your portfolio growing at Y per year Okay that s a little confusing so let s use an example Let s say that your strategy started out with and after years you turned your money into Well what s your cumulative return But what does that tell you about how much you made over the yearsーi e what is the annual return of your portfolio With that we can then easily compare this to things like the SPY which has a CAGR of To do this we annualize our return with the following formula Shoutout Investopedia for a nice and beautiful formula DNow using this formula we can see that our strategy has an annualized return of Nice Now how do we implement this formula As you can see it s pretty straightforward if we have our start and end balance account values def cagr start value float end value float years int return end value start value years cagr account values account values Great Super easy and straightforward When would I use this Use CAGR as a way to easily compare your returns across strategies It s much easier to analyze annualized returns because various strategies will have been run longer than others When you re backtesting you might run one strategy for years whereas another one for These will clearly have different cumulative returns so you will HAVE to use CAGR What is a Good CAGR As mentioned before we use CAGRs to compare with different benchmarks such as the SPY BTC USD Russel etc thus we can now compare our CAGR with these benchmarks and figure out whether or not our model is beating a traditional buy and hold You should also compare this to what you want to make for example real estate returns on average current crypto defi rates such as at Celsius provide upwards of in CAGR Sharpe Ratio Risk Reward Ratio The Sharpe ratio is one of the most renowned ratios out there because it captures two things The amount of reward you are getting i e your returns but also how much risk you are taking on It then takes this and makes it into a metric that can be verbalized as for every unit of reward how much risk am I taking on or put another way what is my reward to risk ratio or another way Am I being rewarded for my risk now with that in mind I think most of you guys can guess that it s going to include some sort of division And indeed it does But the question we have to ask ourselves is well how do we measure reward and risk The ratios use something very different from CAGR Instead of using account values your total asset value at specific times it uses returns or deltas changes in your account value at some times and uses these to ultimately calculate risk the standard deviation in returns and reward the average in returns So why do we use standard deviation of returns Standard deviation if you remember from stats class is a measure of how spread out your data is In the case of returns how spread out your returns is means how volatile your account value is The more spread out your returns are the more likely you are to run into losses and therefore more emotion and therefore more volatility I mean who wants to see their strategy go from up by on one day and then drop by the next day What is Risk Free Rate Our Risk Free Rate is our benchmark for how much we would make if we chose a risk free investment as typically this is the treasury yield or can also be the estimated percentage that you would make if you did a buy and hold or whatever you define as risk free Now what the heck is that square root of n Now we have a list of returns but have no concept of time how do we know if the returns are on a per hour return or on a per day As you can see the various sharpe ratios are going to be different depending on the sample time and sample size don t forget about your stats class and standard error it s ok I kinda also forgot Don t believe me Think about it do stocks move more in a minute or do they move more in a day Every minute a stock might move but every day stocks might move to Well clearly the sharpe ratios are going to be very very different That s why we have to annualize them all so that they are consistent and easily comparable This makes it possible for us to compare strategies that are running on minute data and strategies that run every hour or every day it really doesn t matter Implementation PAY ATTENTION Now to implement this one we ll have to do some manipulation to our account values Let s use the power of numpy to help us out here oh and it s also the same in pandas too We ll be using np diff to take the returns of our account values and resampling them import numpy as npfrom math import sqrt our account values every monthaccount values np array def sharpe account values np array risk free rate annualize coefficient diff np diff account values account values this gets our pct return in the array we ll get the mean and calculate everything we multply the mean of returns by the annualized coefficient and divide by annualized std annualized std diff std sqrt annualize coefficient return diff mean annualize coefficient risk free rate annualized stdsharpe account values Now hold up For the astute observer you might have saw that the sharpe ratio formula doesn t have an annualize coefficient in the numerator but the actual formual does And we also only annualize the denominator Why is that That s because the statistical formula assumes the returns are in the unit of X and not as a ratio i e time This is why in practice we multiply in the numerator and only in the denominator as well This will then give us our correct result What s a Good Sharpe Ratio As we mentioned the sharpe ratio is a ratio of reward vs risk so anything greater than one is phenomenal This means that for every unit of risk you are getting more reward A shapre ratio of is phenomenal a sharpe ratio of even higher well you might be onto something amazing or you might need to check your calculations Great Now let s talk about a similar ratio The Sortino Ratio Sortino RatioThe sortino ratio is almost the exact same as the sharpe ratio except one difference instead of analyzing ALL the standard deviations of returns we only care about the negative returns Why do we do this Well I don t care if my model necessarily makes and then the next they re all wins in my book what I care more about is that I don t want to be losing one day and then the next day I d much rather only lose maximum ImplementationThe implementation is almost the exact same but we ll use some more beautiful numpy magic here to help us out We ll use boolean masking it s actually amazing import numpy as npfrom math import sqrt our account values every monthaccount values np array def sortino account values np array risk free rate annualize coefficient diff np diff account values account values this gets our pct return in the array we ll get the mean and calculate everything remember we re only using the negative returns neg returns diff diff lt annualized std neg returns std sqrt annualize coefficient return diff mean annualize coefficient risk free rate annualized stdsortino account values When would I use Sharpe vs Sortino Again going back to what I said early no one ratio or metric tells the whole story Sharpe and Sortino are great for varying things Sharpe will tell you your overall standard deviation and Sortino will tell you your reward relative to how volatile your downside risk is which induces the most emotion If you care about overall reduced volatility use sharpe if you care about making sure that your model reducing negative downside then use sortino Maximum DrawdownOkay let s stop looking at ratios and focus on another metric Maximum Drawdown The maximum drawdown is as it sounds it s the largest drop between a peak and a trough It shows the largest swing you re expected to have during a trade or strategy and is actually used to calculate the calmar ratio here s the implementation This is important well because we want to know how long our model is going to lose money for or just simply what the biggest loss our strategy incurred over a period of time We can also compare this to our benchmarks and see if we re losing more money and introducing more volatilty than our benchmarks ImplementationFor this implementation let s use pandas because we can use some pandas magic and their functions to calculate the cumulative product of the returns and subsequently the associated peaks How does pandas do this It takes the cumulative product at various points and then finds the maximum peak as it goes making that a series it then takes that and finds the lowest values by using cumulative and finding the ultimate drawdown account values pd Series def max drawdown account values returns account values diff account values cumulative returns cumprod peak cumulative expanding min periods max dd cumulative peak return dd min max drawdown account values What is a good drawdown Drawdowns are always negative so naturally we want less of that negative number so the closer it is to the better Unfortunately we can t ever hope to have a fully drawdown so the closer it is the better typically seeing something that s less than in drawdown is amazing Value At RiskNow Value at Risk is a complicated metric as it s not super super intuitive to visualize as there isn t a direct formula Value at risk VaR is very straightforward what is the maximum capital am I risking at any given point in time in the portfolio i e the average value at risk at any given point in time Think of it as like a sample of the means Thus we use a confidence interval to determine how confidentally we can estimate the value at risk To do this we take your returns and create a normal distribution wow the stats is all coming back now that s great and all But a faster way to do this is to sort all the returns from negative to positive as they will naturally create a normal distribution and take the associated confidence interval Implementationimport numpy as npfrom math import sqrt our account values every monthaccount values np array def value at risk account values alpha float initial value account values returns np diff account values account values returns sorted np sort returns sort our returns should be estimated normal index int alpha len returns sorted return initial value abs returns sorted index value at risk account values Interpreting the ResultsWhat s beautiful about VaR and Conditional VaR is that they are outputted in the unit of the asset so we can directly translate that into something we can directly visualize and see The more value that is at risk the more we should be worried about each trade that is made Thus we want to minimize this as much as possible I mean I want to sleep at night too BetaThe final metric we re going to look at is Beta Now most of you guys have probably heard of alpha it s before in the greek alphabet and it s the amount of returns you have made but beta is something that looks again at something that we care about volatility Specifically beta compares your strategy to a designated benchmark or asset say SPY BTC USD and compares your strategies returns relative to theirs The higher the beta the more volatile your strategy is relative to the base asset Thus the closer the beta is to one the more aligned it is with the base asset and a beta less than one means that your strategy is less volatile which can also mean you can sleep better by using your strategy INSTEAD of using buying the base asset and everyone wants that ImplementationNow implementing this in practice is actually slightly harder because we actually have to get market data There are a multitude of different ways to do this Some work in research environments and non static data others work in liver data but we re going to use one that uses both Blankly Blankly is a package that enables the rapid development of trading algorithms at scale extremely quickly I built an RSI screener bot in lines of code and allows for direct connections with live exchanges for free live trading out of the box Now there is some set up that you have to do but fortunately all you have to do is run a simple command blankly initAnd your entire directory is setup You can check out more info here To actually implement this we re going to make our function a lot more powerful by incorporating market data to directly calculate beta from blankly import Alpaca CoinbasePro supports stocks crypto and forexa Alpaca initialize the Alpaca Exchangeaccount values np array def beta returns baseline SPY return resolution m get the last X month bars of the baseline asset market returns a interface history baseline len returns resolution return resolution m np matrix returns market base returns return np cov m np std market base returns Great now we re hooked up with live data in real time And the beauty of it is that we can easily switch this to use crypto by changing our exchange and that s all we need to do A full function that works across assets Don t want to use blankly for stocks We can also use yfinance and the code would look like this with some restrictions import yfinance as yfaccount values np array def beta returns baseline SPY return resolution m we have to restrict this to get the past year ticker yf Ticker baseline market returns ticker history baseline period y interval m m np matrix returns market base returns return np cov m np std market base returns Why is Beta important Beta is super useful for us because we want to know if our model is gaining a higher reward while also beating the base asset or benchmark i e less volatile This is beautifully paired up with all of our metrics that we ve seen earlier and allows us to get a complete and holistic view of our trading models Let s Sum It Up The Importance of TestingOkay wow that was a lot But I hope that this tutorial was super useful for you to easily see how we can use metrics to make better decisions about our strategies Again make sure you are testing all of your models before deploying them live or putting real money in But also don t ever forget that backtesting can only tell you only a certain amount past performance never defines future success In this article we learned about metrics Sharpe Ratio Reward Risk Ratio across all returnsSortino Ratio Reward Risk Ratio across negative returnsMaximum Drawdown The largest loss during a backtest or strategy periodValue at Risk The value at risk that is made per tradeBeta The volatility of your strategy relative to some benchmark or base asset Always Use Metrics as Comparisons and Not in IsolationRemember that we don t use metrics in isolation to validate our model We have to compare it to a variety of other models strategies and benchmarks How do I easily improve my strategies and these metrics It s hard to say how to best improve every strategy because each strategy is fundamentally different for example strategies that trade on specific assets are going to be limited by the volatility of the underlying asset But I have a couple of recommendations Use Stop LossesStop losses are great ways to make sure that you are not losing money Investor s Business Daily has a great article on this that you can find here on cutting your losses short This will greatly improve your Sortino ratio and sharpe ratios if you re able to adequately exit out of losing trades You are as good as your losses are Make More Frequent TradesThis one might work both well or poorly but the more trades you make the more small wins you will have which will naturally reduce your volatility This might not change your sharpe ratios and sortino ratios but depending on your interval various trades will result in better results This combined with stop losses will also really help Iterate Iterate IterateNothing can beat iteration and rapid optimization Try running things like grid experiments batch optimizations and parameter searches Take a look at various packages like hyperopt or optuna as packages that might be able to help you here Signing Off Here That s all I have I hope you thoroughly enjoyed this post and hopefully this helps you along your quant journey If you want all of the metrics check out this gistIf you still have questions feel free to reach out to me over twitter bfan or join this Discord where we talk about building strategies as much as we breathe If you re excited about this space and quantitative finance we love to hear more about why you re interested and encourage you to get started by building your own trading model using Blankly Cheers to all the quants out there Keep building	2022-01-08 20:41:28
海外TECH	Engadget	NASA finishes deploying the James Webb Space Telescope	https://www.engadget.com/james-webb-space-telescope-deployed-203919497.html?src=rss	NASA finishes deploying the James Webb Space TelescopeNASA is one large step closer to putting the James Webb Space Telescope into service The agency has successfully deployed the JWST s signature gold coated primary mirror completing all major deployments for the instrument The mission crew still has to align the telescope s optics by moving the primary mirror s segments a months long process but it s a strong sign the billion device is in good shape The JWST also requires a third course correction burn as it heads toward the L Lagrange point between the Earth and the Sun Astronomers will use the point to study infrared light without interference potentially offering insights into the early Universe that aren t possible with Hubble and other equipment First images from the telescope won t be available until the summer and it could take much longer before those images translate to meaningful discoveries Even so the deployment is an achievement JWST represents the first time NASA has unpacked a complex observatory in space ーit shows projects like this are viable even if they re unlikely to be commonplace in the near future NASAWebb is fully deployed With the successful deployment amp latching of our last mirror wing that s major deployments complete pins released years of work realized Next to UnfoldTheUniverse traveling out to our orbital destination of Lagrange point pic twitter com mDfmlaszzVーNASA Webb Telescope NASAWebb January	2022-01-08 20:39:19
ビジネス	ダイヤモンド・オンライン - 新着記事	あいおいは「代理店削減不可避」、損保業界の2022年はコスト削減圧力増加でリストラの年に - 総予測2022	https://diamond.jp/articles/-/291209	損害保険	2022-01-09 05:25:00
ビジネス	ダイヤモンド・オンライン - 新着記事	日本生命とメディケア生命が“台風の目”、生保業界の2022年は「成長2大市場」で競争激化 - 総予測2022	https://diamond.jp/articles/-/291208	台風の目	2022-01-09 05:20:00
ビジネス	ダイヤモンド・オンライン - 新着記事	習近平氏の本音は「西側は自滅する」、米中の緊張が2022年も高まる理由 - 総予測2022	https://diamond.jp/articles/-/291207	日本総合研究所	2022-01-09 05:15:00
ビジネス	ダイヤモンド・オンライン - 新着記事	日本製鉄、トヨタに強気の“反撃“で号砲！鉄鋼3社2022年「値上げ」レース - 総予測2022	https://diamond.jp/articles/-/291206	取り組み	2022-01-09 05:10:00
ビジネス	ダイヤモンド・オンライン - 新着記事	みずほの3.11直後2度目の大規模システム障害は必然、前回の教訓を生かせなかった末路 - みずほ「言われたことしかしない銀行」の真相	https://diamond.jp/articles/-/287705	存在意義	2022-01-09 05:05:00
北海道	北海道新聞	＜社説＞道内も感染拡大　医療体制の点検万全に	https://www.hokkaido-np.co.jp/article/631436/	感染拡大	2022-01-09 05:01:00
ビジネス	東洋経済オンライン	日本人が長く見過ごしてきた経済成長の｢犠牲者｣ 資本主義を支えたのはケアワーカーたちだ | リーダーシップ・教養・資格・スキル | 東洋経済オンライン	https://toyokeizai.net/articles/-/500544?utm_source=rss&utm_medium=http&utm_campaign=link_back	新型コロナウイルス	2022-01-09 05:40:00
ビジネス	東洋経済オンライン	ローソン､｢コンビニ大競争｣へ仕込む2つの切り札 社長直下で大改革､人手不足が再燃する懸念も | コンビニ | 東洋経済オンライン	https://toyokeizai.net/articles/-/501214?utm_source=rss&utm_medium=http&utm_campaign=link_back	人手不足	2022-01-09 05:20:00