这一个多月来宏观形势云诡波谲,不可谓不精彩绝伦。还是先说说自己的持仓情况。
具体标的方面,Oil相关减仓约1/3,投入贵金以及债类资产。油价应已经进入平衡价位,基于前期权益类资产和商品类资产价格比值考虑,均值回归模型下如果美油跌至60左右我会考虑再适当加仓。 纳指继续推高,科技龙头和小盘股齐飞,维持QDII定投,小盘股方面GNC这种死狗都开始反弹了,适当减了点补仓。鹅飞了,创造出单支标的盈利记录,真的是没有想到的事情,之前还买了put套保,看不懂,不看好可持续性,已减半仓。德银PB不到0.3,市值不及城商行,佛系地加了点。减仓HK ETF标的,买入put套保,逻辑是金管局跟;不列颠ETF一直半死不活,股息4%+了,继续看好,后期可以持续加仓。其他标的并不明朗,后期适当增配些通胀挂钩债券和短期利率债货币基金。
宏观层面,USD爆发:
美元指数的这一波持续上行始于4月中旬,从4月17日到6月15日,美元指数升值幅度约为6.0%,同期美元对人民币双边汇率升值约2.2%;但端午节后的四个交易日,美元指数的最大升值幅度仅为0.8%,但美元对人民币双边汇率的升值幅度则达到了1.2%。 HKD/RMB也从四月中旬的0.8升值到了0.83.近几天USD指数下跌,RMB并没有像样反弹,反倒补跌。
川怼见完三哥,回到家就向中国加税。之前在财新群里看到有人讲,国内外主流媒体从川怼上台到贸易战开打,前期都做了错误的预判,值得反思。
区域性因素考虑,南美诸国有超跌超卖倾向,待探底稳定后可布局相应ETF长期持有。
最近定投的大A相关基金跌成狗,直接把支付宝年收益拉到负了,坑爹,看了基本标的情况不觉得有太大问题,考虑下收入和现金流,大概减了30%额度,剩下继续坚持,每月也没多少钱。
昨晚放出降准消息,可以说是非常失望了,过去一两年景气周期以来,哪怕是加息25基点,也有逆转民众预期的可能,如今这般面多了放水,水多了加面的局面,继续苟且。等了两年,我是寒心了,重新捡起英语,锻炼好身体,深挖洞广积粮。
言尽于此。
Updated 2018-06-25
Orion要用来做天池比赛的服务器啦,安多一块硬盘用来存雷达数据。放上后需要用parted做分区,然后mkfs格式化,过程和CentOS差不多。这里做一下记录:
lzhenn@thinkstation:~$ sudo parted /dev/sdb
(parted) mklabel GPT
(parted) mkpart primary 0% 100%
(parted) print
Model: ATA HGST HUS724040AL (scsi)
Disk /dev/sdb: 4001GB
Sector size (logical/physical): 512B/4096B
Partition Table: gpt
Disk Flags:
Number Start End Size File system Name Flags
1 1049kB 4001GB 4001GB primary
(parted) quit
lzhenn@thinkstation:~$ partprobe # let the core know the partation
lzhenn@thinkstation:~$ mkfs.ext4 /dev/sdb1 # format
format后,建立fstab并mount -a即可。与CentOS完全一致。
Updated 2018-06-05
In our hierarchy experiments, we hope to make the NO_TOPO series experiments as simple as possible. Thus, it would be great if there is no vegetation on the ground.
In the CLM, we expect there is an external forcing data to control the surface type of ground/vegetation. Then I found this in lnd_in
file
fsurdat = '/users/yangsong3/CESM/input/lnd/clm2/surfdata/surfdata_1.9x2.5_simyr2000_c091005.nc'
If you ncdump
that file, you will see this variable:
double PCT_PFT(lsmpft, lsmlat, lsmlon) ;
PCT_PFT:long_name = "percent plant functional type of gridcell" ;
PCT_PFT:units = "unitless" ;
The first dimension show the corresponding PFT (Plant Functional Types), which define the types and fraction of different landcovers a certain grid (Mosaic method). There is the code table:
Thus, the only thing we need to do is to create a new file with change only in PCT_PFT(0,:,:)=100.0.
Updated 2018-06-05
When we tried to run the model, we got several errors.
surfrd_wtxy_veg_all ERROR: sum(pct) over numpft+1 is not = 100.
490 100.106661159475 0.106661159474527 4473
According to the error, it seems that the sum of pft is not 100. After checking the data, we found several variables:
pct_list=(/"PCT_LAKE", "PCT_WETLAND", "PCT_GLACIER", "PCT_URBAN"/)
So I remain all the pct about water, urban, and glacier, and make the bare ground equal to 100-sum(pct_list)
.
Note that after changing all the landuse pct, the default CLM initial condition cannot satisfy the model, so we need to “cold” start the model. Use env_run.xml
to set it.
<entry id="CLM_FORCE_COLDSTART" value="on" />
Updated 2018-06-06
Some issues happened after 10+ years integration. In no_topo experiment, the model exit with error in 0013-03, with Error message:
(shr_sys_abort) ERROR: remap transport: bad departure points
(shr_sys_abort) WARNING: calling shr_mpi_abort() and stopping
Warning: Departure points out of bounds in remap
my_task, i, j = 31 20 13
application called MPI_Abort(MPI_COMM_WORLD, 1001) - process 0
dpx, dpy = -28367.4880120091 54137.9986747345
HTN(i,j), HTN(i+1,j) = 30148.1831506031 30030.4921250813
HTE(i,j), HTE(i,j+1) = 46470.5086842257 46750.9484318852
istep1, my_task, iblk = 1310809 31 2
Global block: 241
Global i and j: 19 372
After searching, it seems that the CICE model encountered a CFL violation problem:
This is essentially a CFL violation in the CICE model. You would need to reduce the timestep. In the CESM, the CICE timestep is tied to the coupling interval with the atmosphere. Have a look at ATM_NCPL and ICE_NCPL. Try doubling ATM_NCPL and this should help. For high resolution (0.1-degree) you can use xndt_dyn or ndtd depending on the version of the code. This allows you to take multiple dynamic timesteps per thermodynamic timestep.
I found ATM_NCPL in env_run.xml:
<!--"base period associated with NCPL coupling frequency, valid values: hour,day,year,decade (char) " -->
<entry id="NCPL_BASE_PERIOD" value="day" />
<!--"number of atm coupling intervals per NCPL_BASE_PERIOD (integer) (char) " -->
<entry id="ATM_NCPL" value="48" />
Since the atm run in dtime=1800s
, double the ATM_NCPL
is meaningless.
Another strategy is to double the CICE dynamic timesteps per thermodynamic timestep.
The dynamics timestep should be reduced to integrate past this problem. Set
xndt_dyn = 2
in the namelist and restart the model. When the job completes set the value back to 1.
That is quite interesting, set it back when pass the unstable point? How could there be this operation?!
Still confused, then I turned to NO_TOPO_SCON
and found another similar error happened in the ocean model:
POP Exiting...
POP_SolversChronGear: solver not converged
POP_SolverRun: error in ChronGear
POP_BarotropicDriver: error in solver
Step: error in barotropic
And I searched and found the POP FAQ:
If your case has been running successfully for a period of time, but then the ocean model suddenly fails to converge, you might have encountered a time-stepping instability. If so, you will need to decrease the ocean timestep and restart your case. We recommend you back up at least one month prior to the time of nonconvergence, decrease the ocean-model timestep by about 15-20%, and then restart your case.
And the method to change the timestep:
To change the size of the POP2 timestep, you simply change the value of dt_count in your $CASE/user_nl_pop2 file.
To decrease the POP2 timestep, you must increase the POP2 time_manager_nml namelist variable dt_count.
For example, if the ocean grid in your case is gx1v6, then the default value for dt_count is 23. In order decrease the timestep by roughly 15%, you would increase dt_count to 26. To do this, put the following line in your
$CASE/user_nl_pop2
file:
dt_count = 26
You do not need to rebuild the executable in order to change dt_count.
Considering doubling the CICE dynamic timesteps will significantly increase the integration time. I tried the following method:
user_nl_pop2
:dt_count=26
According to the paleo FAQ:
However, if you have changed your land cover in your new simulation (for example, if you have changed your land ice distribution), you will first need to create a new clm2.r file that conforms to the new land cover assignments in your modified surface datafile. Creating a new clm2.r file is a two step process. Step 1: Run a short (e.g., 5 day) startup simulation with arbitrary initial conditions and your new surface dataset. The 5 day simulation will produce a restart file that is consistent with the ice distribution in your new surface dataset, which is different from the original experiment that you wish to branch from. Use the tool, interpinic, to re-map the restart data from the original simulation to the new restart file created by the 5-day run.
I used to cold start the model, it seems this can be a big problem:
CLM4 does not require an initial condition file and can be initialized with arbitrary initialization (finidat=” “). However, arbitrary initialization starts the model from essentially bare ground, with vegetation, soil carbon, soil nitrogen, and soil moisture set to zero. The land model will spin up all variables at once, but will therefore require many hundreds of years to come into equilibrium.
Following the instruction, we first startup a NO_TOPO
experiment with 5 day integration and our own bare ground surface data. Then, use interpinic
to remap a REAL_WORLD
restart file to our 5-day restart file:
Path: /users/yangsong3/CESM/cesm1_2_2/models/lnd/clm/tools/clm4_0/interpinic
./interpinic -i /users/yangsong3/L_Zealot/B/B20f19-realworld/exe/B20f19-realworld.clm2.r.0021-01-01-00000.nc -o /users/yangsong3/L_Zealot/B/B20f19-topo/clm.r/B20f19-topo.clm2.r.interpinic.nc
After that, I thought for a while and tried to comment out the pop timestep change, turn off CLM cold start, and use frequent restart file output (10 year).
Updated 2018-06-11