1. BAYESIAN NETWORK MODELS OF PORTFOLIO RISK AND RETURN

1. BAYESIAN NETWORK MODELS OF PORTFOLIO RISK AND RETURN
Catherine Shenoy and Prakash P. Shenoy
A Bayesian network is a tool for modeling large multivariate probability models and for making inferences from such models. A Bayesian network combines traditional quantitative analysis with expert judgement in an intuitive, graphical representation. In this paper, we show how to use Bayesian networks to model portfolio risk and return. Traditional financial models emphasize the historical relationship between portfolio return and market return. In practice, to forecast portfolio return, financial analysts include expert subjective judgement about other factors that may affect the portfolio. These judgmental factors include special knowledge about the stocks in the portfolio that is not captured in the historical quantitative analysis. We show how a Bayesian network can be used to represent a traditional financial model of portfolio return. Then we show how expert subjective judgement can be included in the Bayesian network model. The output of the model is the posterior marginal probability distribution of the portfolio return. This posterior return distribution can be used to obtain expected return, return variance, and value-at-risk.
The main goal of this paper is to show how Bayesian networks can be used to model portfolio risk and return. Bayesian networks have been used as a tool for modeling large multivariate probability models and for making inferences from such models (Pearl 1986, Lauritzen and Spiegelhalter 1988, Shenoy and Shafer 1990). A Bayesian network combines traditional quantitative analysis with an analyst’s judgement in an intuitive, graphical representation. It allows an analyst to visualize the relationships among the variables in the model. Finance models focus on the historical, quantitative relationships between economic variables. However, financial analysts usually combine historical data with qualitative information and judge how this information affects stock returns, market return, interest rates, or any other input to a portfolio model. For example, the anti-trust lawsuit against Microsoft affects the stock returns of many companies, but this type of information is difficult to incorporate in traditional return models. Bayesian networks are especially well suited for situations that combine quantitative and qualitative information. In this paper we provide an overview of how to combine traditional financial models with judgments about qualitative information in a Bayesian network framework. Traditional portfolio return models are static. There is no systematic way to update results in the light of new information. A Bayesian network representation of portfolio return allows analysts to incorporate new information, to see the effect of that information on the return distributions for the whole network, and to visualize the distribution of returns, not just the summary statistics. In a Bayesian network an analyst first determines the qualitative structure of the model in an intuitive graphical way. A traditional portfolio model can easily be represented as a Bayesian network. From that basic structure, quantitative information is then added to the model. Any change to either the qualitative or quantitative structure of the model is immediately reflected as the model is updated. The output of the Bayesian network analysis is a distribution of portfolio returns based on the qualitative and quantitative structure of the model. Most traditional financial models rely on strong implicit assumptions about the independence of various factors incorporated in the model. In a Bayesian network model, the analyst can explicitly model the dependence or independence of the factors. It is then possible to determine the sensitivity of the portfolio return to those simplifying assumptions by relaxing the assumptions. Portfolio risk analysis is typically based on the assumption that the securities in the portfolio are well diversified. Portfolios that contain securities with several correlated risk factors do not meet the well-diversified criteria. Some portfolios by construction contain a predominant factor. Examples include sector or regional mutual funds. Other portfolios may be constrained in their ability to diversify. Examples include financial institutions’ loan portfolios or an individual’s personal portfolio. Using a Bayesian network model, we can examine the effect of risk

2
C. Shenoy and P.P. Shenoy
concentration on portfolio risk. From the posterior return distribution, we calculate the portfolio variance and compare it to a non-diversifiable risk measure. Using this approach, we explore the dependence and independence assumptions used in traditional portfolio models. Traditional financial analysis focuses on summary statistics — expected returns, beta, variance or standard deviation of returns. Recently value-at-risk analysis has emphasized consideration of the whole distribution of returns, or at least, the left-hand tail of the distribution. Since the output of the Bayesian network model is a posterior portfolio return distribution, we can also calculate the cutoff return for a value-at-risk calculation. As information is added to the network, the return distribution in the network reflects those changes and the cutoff value-at-risk is updated as well. The rest of the paper proceeds as follows. In section two, we briefly discuss some traditional portfolio return models. In section three, we define and describe the semantics of a Bayesian network. In section four, we model a simple gold stock portfolio using a Bayesian network. Finally, in section five we discuss some modeling issues and limitations of Bayesian networks. 1.1. Traditional Portfolio Return Models In a traditional portfolio analysis, the hypothesized relationship assumes that the rate of return on an asset is a linear function of the market rate of return and an asset specific factor, as follows: Ri = ai + bi RM + Ei (1)
where Ri denotes return on asset i, ai and bi are constants, RM denotes return on a market index, and Ei denotes an uncertain variable related to asset-specific factors. Many studies have shown that some specific identifiable components of risk are not fully accounted for by just a market index. These studies have found that industry-specific risk, countryspecific risk, and many other components account for correlation among individual securities. King (1966) finds that market factors explain 30% of return variation and industry factors explain an additional 10%. Goodman (1981) shows that in country-specific diversified portfolios, significant mis-measurement of risk occurs if the market proxy does not include global factors. The arbitrage pricing theory (APT) (Ross 1976) and other multi-factor models (see Elton and Gruber 1997) extends the single factor model to account for these additional, identifiable factors. The multi-factor model can be represented as an expanded version of equation (1.1): Ri = ai + bi1 F1 + bi2 F2 + bik Fk + Ei where F1, …, Fk denote the k independent factors, and bi1, …, bik are constants. Portfolio return, denoted by Rp, is defined as the weighted average of the individual returns that comprise the portfolio, R p = ? w i R i . Where wi denotes the proportional amount invested in
i =1 n
(2)
security I, and n denotes the number of securities in the portfolio. Portfolio variance, denoted by sp, is given by:
2 2 2 s 2 = b1P s 2 +…+ b 2 s 2 + w1 s 2 + … + w1 s 2 + F E p kP F E
1 1 1 n
??w w
i j=1 i =1
n
n
j
cov E i , E j
(
)
k
(3)
where bkp denotes the portfolio beta for the kth independent factor, s 2 denotes the variance of the F kth independent factor, and s 2 denotes the residual asset-specific variance. E
i
It is assumed that all asset-specific uncertain variables, E1, …, Ek, are mutually independent. So portfolio variance simplifies to:
2 2 s 2 = b1P s 2 + o + b 2 s 2 + w1 s 2 + K + w 2 s 2 P F kP Fk E n E
1 1 n
(4)

Bayesian Network Models of Portfolio Risk and Return
2 Portfolio risk is divided into two components — diversifiable risk, w1 s 2 + K + w 2 s 2 , and E n E 2 2 2 2 non-diversifiable risk, b1P s F + o + b kP s Fk . It is normally assumed that diversifiable risk is small since each w 2 is small. However, in study of bank loan portfolios, Chirinko and Guill (1990) find i that assuming the covariance terms are zero leads to portfolio variances being under-estimated from 24.6% to 45.75%. For an equally weighted loan portfolio with 46 industries, the variance was underestimated by 36.36%.
1 n 1
3
1.2. Bayesian Networks Bayesian networks have their roots in attempts to represent expert knowledge in domains where expert knowledge is uncertain, ambiguous, and/or incomplete. Bayesian networks are based on probability theory. A Bayesian network model is represented at two levels, qualitative and quantitative. At the qualitative level, we have a directed acyclic graph in which nodes represent variables and directed arcs describe the conditional independence relations embedded in the model. Figure 1 shows a Bayesian network consisting of four discrete variables: Interest Rate (IR), Stock Market (SM), Oil Industry (OI), and Oil Company Stock Price (SP). At the quantitative level, we specify conditional probability distributions for each variable in the network. Each variable has a set of possible values called its state space that consists of mutually exclusive and exhaustive values of the variable. In Figure 1, e.g., Interest Rate has two states: ‘high’ and ‘low;’ Market has two states: ‘good’ and ‘bad;’ Oil Industry has two states: ‘good’ and ‘bad;’ and Oil Company Stock Price has two states: ‘high’ and ‘low.’ If there is an arc pointing from X to Y, we say X is a parent of Y. For each variable, we need to specify a table of conditional probability distributions, one for each configuration of states of its parents. Figure 1 shows these tables of conditional distributions—P(IR), P(SM | IR), P(OI), and P(SP | SM, OI).
1.2.1.
Semantics of Bayesian Networks
A fundamental assumption of a Bayesian network is that when we multiply the conditionals for each variable, we get the joint probability distribution for all variables in the network. In Figure 1, e.g., we are assuming that P(IR, SM, OI, SP) = P(IR) ? P(SM | IR) ? P(OI) ? P(SP | SM, OI), where ? denotes pointwise multiplication of tables. The rule of total probability tells us that P(IR, SM, OI, SP) = P(IR) ? P(SM | IR) ? P(OI | IR, SM) ? P(SP | IR, SM, OI). Comparing the two, we notice that we are making the following assumptions: P(OI| IR, SM) = P(OI), i.e., OI is independent of IR and SM; and P(SP | IR, SM, OI) = P(SP | SM, OI), i.e., SP is conditionally independent of IR given SM and OI. Notice that we can read these conditional independence assumptions directly from the graphical structure of the Bayesian network as follows. Suppose we pick a sequence of the variables in a Bayesian network such that for all directed arcs in the network, the variable at the tail of each arc precedes the variable at the head of the arc in the sequence. Since the directed graph is acyclic, there always exists one such sequence. In Figure 2 one such sequence is IR SM OI SP. The conditional independence assumptions in a Bayesian network can be stated as follows. For each variable in the sequence, we assume that it is conditionally independent of its predecessors in the sequence given its parents. The key point here is that missing arcs (from a node to its successors in the sequence) signify conditional independence assumptions. Thus the lack of an arc from IR to OI signifies that OI is independent of IR; the lack of an arc from SM to OI signifies that OI is independent of SM; and the lack of an arc from IR to SP signifies that SP is conditionally independent of IR given SM and OI.

4
C. Shenoy and P.P. Shenoy
Figure 1. A Bayesian Network with Conditional Probability Tables
P(IR) IR: Interest Rate P(SM | IR) Good Bad High 0.20 0.80 Low 0.70 0.30 SM: Stock Market High 0.25 Low 0.75
P(OI)
Good 0.40
Bad 0.60
OI: Oil Industry
SP: Oil Company Stock Price
P(SP | SM, OI) High Low Good, Good 0.80 0.20 Good, Bad 0.60 0.40 Bad, Good 0.50 0.50 Bad, Bad 0.10 0.90
In general, there may be several sequences consistent with the arcs in a Bayesian network. In such cases, the lists of conditional independence assumptions (associated with each sequence) are equivalent using the laws of conditional independence (Dawid 1979). There are other equivalent graphical methods for identifying conditional independence assumptions embedded in a Bayesian network graph (see Pearl (1988) and Lauritzen et al. (1990) for examples.). 1.1.2. Making Inferences in Bayesian Networks
Once a Bayesian network is constructed, it can be used to make inferences about the variables in the model. The conditionals given in Bayesian network representation specify the prior joint distribution of the variables. If we observe (or learn about) the values of some variables, then such observations can be represented by tables where we assign 1 for the observed values and 0 for the unobserved values. Then the product of all tables (conditionals and observations) gives the (unnormalized) posterior joint distribution of the variables. Thus the joint distribution of variables changes each time we learn new information about the variables. In theory, the posterior marginal probability of a variable X, say P(X), can be computed from the joint probability by summing out all other variables except X one by one. In practice, such a naiv e approach is not computationally tractable when we have a large number of variables because the joint distribution has an exponential number of states and values. The key to efficient inference lies in the concept of local computation where we compute the marginal of the joint without actually computing the joint distribution. A key feature of a Bayesian network is that it describes a joint distribution from the local relationships—such as a node and its parents. Instead of tackling the whole collection of variables simultaneously, Bayesian networks use the concept of factorization. Factorization involves breaking down the joint probability distributions into subgroups called factors in such a way that the naive computations described above need only be performed within each subgroup. Since the state space of a subgroup is much smaller than that of the joint probability distribution, the calculations become manageable.

Bayesian Network Models of Portfolio Risk and Return
5
Bayesian networks can be used for two types of inference.1 Often we are interested in the values of some target variables. In this case, we make inferences by computing the marginal of the posterior joint distribution for the variables of interest. Consider the situation described by the Bayesian network in Figure 1. Suppose we are interested in the true state of Oil Company Stock Price (SP). Given the prior model (as per the probability tables shown in Figure 1), the marginal distribution of SP is 0.502 for high and 0.498 for low. Now suppose we learn that Interest Rate is low. The posterior marginal distribution of SP changes to 0.554 for high and 0.446 for low. Suppose we further learn that the state of Oil Industry is good. Then the marginal distribution of SP changes to .71 for high and 0.29 for low. This type of inference is referred to as ‘s u m propagation.’ Sometimes we are more interested in the configuration of all variables (“the big picture”) rather than the values of individual variables. In this case, we can make inferences by computing the mode of the posterior joint distribution, i.e., a configuration of variables that has the maximum probability. Consider again the situation described by the Bayesian network in Figure 1. Given the prior model (as per the probability tables shown in Figure 1), the mode of the prior joint distribution is (low interest rate, good stock market, bad oil industry, low oil company stock price). Now suppose we learn that Interest Rate is high. The mode of the posterior joint distribution changes to (high interest rate, bad stock market, bad oil industry, high oil company stock price). This type of inference is referred to as ‘max propagation.’ The results of inference are more sensitive to the qualitative structure of the Bayesian network than the numerical probabilities (Darwiche and Goldszmidt 1994). For decision making, the inference results are robust with respect to the numerical probabilities (Henrion et al. 1994). There are several commercial software tools such as Hugin (https://www.wendangku.net/doc/af3807460.html,) and Netica (https://www.wendangku.net/doc/af3807460.html,) that automate the process of inference. These tools allow the user to enter the Bayesian network structure graphically, enter the numerical details, enter any additional information, and then do inference of either type. The results of the inference are then shown graphically using bar charts. 1.3. A Bayesian Network Model of Multi-Factor Portfolio Return In this section, we first describe a traditional security return model as a Bayesian network. Then we demonstrate how some of the independence assumptions in a traditional security return model can be relaxed using Bayesian networks. 1.3.1. Description of a Gold Stock Portfolio Network
A security return model is a conditional expectations model and is usually estimated using least squares regression, so that
E ( R i | Fi , F2 , o Fk ) = a i + b i1 F1 + o + b ik Fk
(5)
We can easily regard this as a Bayesian network model where the factors F1, …, Fk are regarded as mutually independent variables. The following Bayesian network model considers the return on an equally weighted portfolio of three stocks from the gold mining industry. Their ticker symbols are BGO, ABX, and AEM. In the representation shown in Figure 2, market and gold returns are parents of each stock’s return. In addition, each stock return has an idiosyncratic component. The portfolio return is a function of each stock return and the weight of each stock in the portfolio. In the graph, each random variable is shown as a node. A variable that is conditionally deterministic given the values of its parents is shown as a double-bordered node.
1
Lauritzen and Speigelhalter (1988), Jensen et al. (1990) and Shenoy and Shafer (1990) have devised propagation algorithms to perform efficient probabilistic inference.

6
C. Shenoy and P.P. Shenoy
Figure 2. A Bayesian Network Model for a Portfolio
BGO Effects
Market
BGO
ABX Portfolio Return ABX Effects
Gold
AEM
AEM Effects
The graphical model is supplemented by numeric information about the conditional probability distributions. For each variable in the model, we define the conditional probability distributions given each combination of states in the parent nodes. In Figure 1, market, gold, BGO effects (eBGO), ABX effects (eABX), and AEM effects (eAEM ) have no parents, so we specify a prior distribution for these nodes. BGO has market, gold and BGO effects as its parents. Since BGO is a conditionally deterministic node, we specify a functional relation for its value as a function of the states of its parents. We also define a similar relation for ABX and AEM. Finally, Portfolio Return has BGO, ABX, and AEM. Since Portfolio Return is also a conditionally deterministic variable, it has a unique deterministic state given by some functional relation. 1.3.2. Inputs for the Bayesian Network
The conditional relationship for each of the stocks can be specified as an equation, such as the multi-factor model specified in equation (1.2); as a discrete conditional probability table; or as a continuous conditional probability distribution. Any combination of historical data, forecasts, expert knowledge, or beliefs can be used to estimate the conditional relationships. Initially for the primary inputs market, gold, and the individual stock effects we do not specify an explicit conditional relationship. We specify the a priori distribution for each of these as a normal distribution with parameters estimated over an arbitrary period from January 1996 through February 1998. We estimate weekly returns for these inputs. For example, using historical data over the estimation period, weekly market return has mean 0.55% and standard deviation 2.28%. These inputs are summarized in Table 1 below. For each stock effect, we assume a mean of zero and a standard deviation equal to the standard error of the regression equation. For each of the stock returns, the distribution is conditioned on the market and gold returns and a stock-specific effect. The conditional distribution for each stock node is normal with mean based on equation (1.2) with factors of market and gold returns. The mean stock return is the

Bayesian Network Models of Portfolio Risk and Return
7
estimated regression for each stock node. The standard error of the regression is the standard deviation of the distribution. The regression estimates are based on weekly returns from January 1996 through February 1998. The result of the equation using the state values of the inputs determines the stock return node. For ABX the conditional relationship is the estimated regression equation: ABX return = 0.17 + 0.366*market + 2.26*gold + eABX, (6)
A summary of all of the inputs and regression coefficients used for each of the nodes is presented in Table 1. Table 1. Parameter and Regression Estimates Used in Portfolio Network
BGO Bema Gold Corp. 0.68 12.55 1.48 0.26 3.96 11.27 ABX Barrick Gold Corp. –0.11 5.00 0.17 0.37 2.26 3.63 AEM Agnico Eagle Mines –0.42 6.58 0.05 0.27 2.76 5.27 Market S&P 500 Index 0.55 2.28 Gold London PM Gold Fix –0.23 1.43
Description Average monthly return (%) Standard deviation (%) Regression estimates: Intercept Market coefficient Gold coefficient standard error
In order to generate the conditional return distributions for BGO, ABX, AEM, and the portfolio, we use Monte Carlo simulation. In the simulation, we specify the functional relationships between the nodes to generate the estimated return distributions. The portfolio return distribution is a simple average of the stock returns; that is, an equally weighted portfolio. Table 2 reports additional statistics for each of the conditional probability return distributions based on the simulation results. The average weekly portfolio return is 0.06% with a standard deviation of 6.15%. Table 2. Conditional Probability Distributions for Bayesian Network Mean standard deviation Minimum Maximum 5th percentile 50th percentile 95th percentile Market 0.55 2.28 –6.94 8.69 –3.19 0.55 4.30 Gold –0.23 1.43 –5.10 4.48 –2.57 –0.23 2.12 BGO 0.72 12.73 –48.86 49.73 –20.00 0.82 20.61 ABX –0.43 6.60 –24.84 20.15 –11.34 –0.54 10.13 AEM –0.15 4.95 –17.46 18.64 –8.09 –0.26 8.15 Portfolio 0.06 6.15 –19.39 18.52 –10.32 0.18 9.85

8
C. Shenoy and P.P. Shenoy
Figure 3. Portfolio Return Statistics from Bayesian Network Model
Descriptive Statistics
Variable: PortRet
Anderson-Darling Normality Test A-Squared: 0.25 p-value: 0.74 Mean Std Dev Variance Skewness Kurtosis n of data Minimum 1st Quartile Median 3rd Quartile Maximum 0.05 6.17 38.10 0.09 -0.06 1000.00 -18.07 -4.20 -0.10 4.34 21.74
-20
-10
0
10
20
95% Confidence Interval for Mu
95% Confidence Interval for Mu -0.33 0.43 -0.5 0.0 0.5 95% Confidence Interval for Sigma 5.91 6.46
95% Confidence Interval for Median
95% Confidence Interval for Median -0.52 0.40
Figure 3 shows the simulated distribution of portfolio returns from the Bayesian network based on 10,000 iterations. Because we modeled the inputs to the network as normal distributions, the portfolio return distribution is also approximately normal. From the confidence interval for the mean and the median, we see that the mean return on this portfolio is not significantly different from zero. In Section 3.4 we compare this model and several other Bayesian network models to the actual portfolio return. 1.3.3. Bayesian Network as a Management Tool
A principal advantage of a Bayesian network representation of portfolio risk is in its flexibility as a management tool. In this section we show how new evidence can be entered into a network and how new information can be added to a network. Studies (see Henrion et al. (1994, 1996), and Pradhan et al. (1996)) have shown that the graphical representation of the conditional probabilities is the most important step in modeling. The exact numerical form is of secondary importance. Most decisions will be robust as long as the conditional independence relationships as encoded in the network are specified correctly. Managers usually have a good idea of the influences on a portfolio, but not their exact numerical form. New information can be easily incorporated in the model. For example, suppose we learn that BGO will perform well in the next period if a new product is launched, but BGO will remain flat if the product is not launched on time. We also believe that it is very likely that the product launch will be on time. Specifically we model the evidence as a table of likelihoods as shown in Table 3 below where a return of 10 is four times more likely than a return of 0. Table 3. Evidence for BGO State for BGO Likelihoods Return of 0 0.20 Return of 10 0.80 We incorporate this new evidence in the model, and recompute the marginals of the posterior distributions (as shown in Table 4) to reflect the new information.

Bayesian Network Models of Portfolio Risk and Return
9
Table 4. Revised Conditional Return Distributions Mean standard deviation Minimum Maximum 5th percentile 50th percentile 95th percentile BGO 4.30 4.95 0.00 10.00 0.00 0.00 10.00 Portfolio Return 1.25 3.69 –12.06 12.66 –4.73 1.09 7.27
A Bayesian network can also accommodate some types of information that is not easily incorporated in other types of models. Suppose an analyst learns that AEM and ABX have a common supplier whose favorable actions will affect both AEM and ABX. This new information can be added to the network using subjective probabilities. Figure 4 shows the addition of a new node that directly affects the nodes ABX effects (eABX) and AEM effects (eAEM). Figure 4. Revised Bayesian Network with Supplier Information
BGO Effects
Market
BGO
ABX Portfolio Return ABX Effects
Gold
AEM
Supplier
AEM Effects
1.3.4.
Additional Conditions in the Portfolio Return Model
The model we specify in Figure 2 is based on the traditional finance model that assumes the residual correlations are independent. We use the original return data to calculate the regression residuals and find the residual correlations among the stocks. The residual correlation is reported in Table 5. We find relatively large residual correlation between eABX and eAEM and between eBGO and eAEM. We use this correlation data to specify three additional Bayesian network models that take into account these dependencies.

10
C. Shenoy and P.P. Shenoy
Table 5. Residual Correlations BGO Effects ABX Effects AEM Effects BGO Effects 1 0.1814 0.3367 ABX Effects 1 0.4427 AEM Effects
1
This high correlation indicates that there are still unmodeled factors which affect these stocks. If these factors are unknown, it is still possible to model the dependencies among the residuals. For any ordering of eBGO, eAEM, and eABX, using the multiplication rule, we have f(eBGO, eAEM, eABX) = f(eBGO) f(eAEM
| eBGO) f(eABX | eBGO, eAEM).
(7)
Any other ordering of the residuals are equivalent to this ordering. This fact implies that a Bayesian network that captures all residual correlation can be specified with the following representation: Figure 5. A Bayesian Network Model with Correlated Residuals
BGO Effects
Market
BGO
ABX Portfolio Return ABX Effects
Gold
AEM
AEM Effects
If X and Y are bivariate normal, the conditional distribution f(y | x) is given as:
ê rs ? f ( y | x) ~ Ná y (x - m x ), (1 - r2 )s 2 ? . y ? sx ˉ
(8)

Bayesian Network Models of Portfolio Risk and Return
11
So if we assume that eBGO , eABX , and eAEM are multivariate normal, we specify three additional models with the following independence conditions: 2 1. Model 2: eBGO?eAEM, i.e., eAEM and eBGO are independent of eABX; 2. Model 3: eAEM?eABX, i.e, eAEM and eABX are independent of eBGO; and 3 . Model 4: eB G O?eAEM?eABX, i.e., eBGO and eA B X are conditionally independent given eAEM. To see the effect of adding these conditional probabilities we compare the actual returns over the three year period to each of the Bayesian network models and to the finance model. In Table 6, we find that each model slightly overestimates the mean and underestimates the standard deviation. Table 6. Comparison of Actual Portfolio Return and Portfolio Return Models
% difference from actual Mean Std 5.4 -1.8 5.1 4.1 7.3 -35.9 -8.8 -2.9 -8.8 -0.4
Portfolio Actual portfolio return Finance Model – Non-diversifiable risk 1. Simple Bayes Net (BN) 2. BN with eBGO?eAEM 3. BN with eAEM?eABX 4. BN with eBGO?eAEM?eABX
Mean 0.050 0.053 0.049 0.053 0.052 0.054
Std(Risk) 6.8 4.3 6.2 6.6 6.2 6.7
Actual portfolio return is measured assuming weekly portfolio rebalancing. The mean of the actual portfolio return is the arithematic average of the weekly portfolio returns. For the finance model we measure non-diversifiable risk as b 2 ,m s 2 + b 2 ,g s 2 , where b 2 ,m is the market p m p g p portfolio beta squared, and b 2 ,g is the gold portfolio beta squared. The portfolio beta is the weighted p average of the appropriate gold or market return coefficients from the three regression equations. We see that modeling the effects with the largest residual correlations, rBGO,AEM and rABX,AEM provides improvements in the risk and return estimation Table 7 reports confidence intervals for the mean and standard deviation for the four Bayesian network models and the actual return. There are 115 weekly observations for the actual returns. The Bayesian network models are based on simulations of 10,000 iterations.
2
Model 1 is the Bayesian network shown in Figure 2 where we are assuming that the three effects are mutually independent.

12
C. Shenoy and P.P. Shenoy Table 7. 95% Confidence Intervals for m and s Conf. Interval for m Lower Upper –1.200 1.300 –0.334 0.432 –0.355 0.460 –0.331 0.435 –0.365 0.472 s 6.8 6.2 6.6 6.2 6.7 Conf. Interval for s Lower Upper 6.1 7.6 6.0 6.4 6.3 6.8 6.0 6.4 6.1 7.6
Actual Model 1 Model 2 Model 3 Model 4
x 0.050 0.049 0.053 0.052 0.054
1.4. Modeling Issues and Limitations Bayesian networks are able to incorporate different types of information. In this section we address the issues of how to find inputs to the network, discrete vs. continuous probability distributions, and some model limitations. 1.4.1. Inputs to the network
Two types of inputs are needed for the network. First, factors that affect each asset return in the portfolio have to be identified. Then the conditional probability distributions for the asset returns have to be specified. Both types of inputs can be any combination of empirical data, expectations, judgment, or forecasts. Many empirical studies have attempted to identify the factors that cause variation in security returns. Roll and Ross (1980), Dhrymes et al. (1984), Chen et al. (1986), Elton and Gruber (1997) identify factors such changes in inflation, industrial production, and yield spreads. Other portfolio specific factors may also be important. For example, a geographically limited portfolio would have a factor relating regional economic conditions to the stock returns. Common production inputs, customers, and other special circumstances can also be included. Empirical analysis tools such as linear regression, factor analysis, time series analysis, neural nets and data mining techniques can all be used to generate the conditional probabilities for the dependent nodes. These tools examine the historical relationship among the nodes. Using these analytical tools may be equivalent to using current financial models, if the independence assumptions are the same. For example, Model 1 is equivalent to a traditional multi-factor model because each specific stock effect is assumed to be mutually independent, and all inputs are based on historical data. Judgment and forecasts can be added to the model by revising the conditional probability tables for the nodes, by revising the priors for nodes with no parents, or by adding new nodes. Studies (see Henrion et al. 1994, 1996), and Pradhan et al. (1996)) have shown that the graphical representation of the probability model is the most important step in modeling. The numerical details of the probability model are of secondary importance. Most decisions will be robust as long as the conditional independence relationships as encoded in the network are specified correctly. Managers usually have a good idea of the influences on a portfolio, but not their exact numerical form. Discrete conditional distributions can be used as approximations of the exact form of the distribution. The Bayesian network representation forces the modeler to make explicit judgments of the causal structure of the model. Traditional statistical models have an implicit causal structure that is not always appropriate. The decision-maker can examine the effect of assuming independent residuals and other factors in the model. For example, a model may have a geographic factor and a market factor. In a multi-factor model it is usually assumed the factors are independent; however, the market and geographic factor may not be independent. In a Bayesian network the dependence between the two factors can also be modeled.

Bayesian Network Models of Portfolio Risk and Return
13
1.4.2.
Value-at-Risk
The value at risk (VAR) for a portfolio is the expected maximum loss over a target horizon within a given confidence interval. Recent, large corporate losses in Orange County, Barings Bank, Daiwa and others have received media and regulatory attention. Academicians, regulators, and financial managers have asserted to the need for a better method of summarizing risk. The value at risk metric is one way to quantify portfolio risk. This measure is easily incorporated into a Bayesian network model. The VAR measure is a linear function of the portfolio return defined as: VAR = W0 R*, (9)
where W0 is initial investment and R* is the cutoff portfolio return for the ith percentile. Using Model 1 at a cutoff percentile of 5, R* and is equal to –10.32 percent, and VAR = W0(-0.1032) for a period of one week. An important consideration for VAR measures is the time over which risk is measured. 1.1.3. Limitations of Bayesian Network Models
A major limitation in using Bayesian networks to model portfolio returns is determining the graphical structure of the Bayesian network model. A graphical structure can either be obtained subjectively from an expert or one can be induced from data. The latter technique is the subject of current research in the uncertain reasoning literature (see (Heckerman 1997) for a recent survey). Once a graphical structure is obtained, determining the numerical parameters of the model is straightforward when securities are publicly traded and when data is often readily available. If all variables in a Bayesian network are discrete, then the marginal distribution of any variable can be computed exactly using the local computational algorithms proposed, e.g., by Pearl (1988), Lauritzen and Spiegelhalter (1988), and Shenoy and Shafer (1990). These algorithms are encoded in commercial software such as Netica (https://www.wendangku.net/doc/af3807460.html,) and Hugin (https://www.wendangku.net/doc/af3807460.html,). Since security returns are usually modeled as continuous variables, exact computation of the Bayesian network is not always possible. We can either discretize the distributions or use simulation methods. We can compute the posterior marginals approximately using Monte Carlo methods (see, e.g., Henrion 1988). As the number of variables grows, even Monte Carlo methods require an inordinate amount of sampling for a decent approximation. In such cases, Markov Chain Monte Carlo methods have been proposed for faster convergence (see, e.g., Gilks et al. 1996). In our example, we assumed an equally weighted portfolio. If a manager is evaluating a currently held portfolio, the weights will change as the stock prices of the component returns change. Therefore assuming constant weights implies constant rebalancing of the portfolio (to maintain the constant weights) as the stock prices change. Of course, it is not possible to rebalance a portfolio without incurring transaction costs, so the actual return from this type of a portfolio would be lower. It is possible to construct a Bayesian network that calculates portfolio return based on share prices and constant number of shares held. However, such a model is quite different from a traditional finance model, and is the subject of future research. 1.5. Conclusions The main goal of this paper is to propose Bayesian networks as a tool for modeling portfolio returns. Bayesian networks allow us to explicitly model the dependence between the various factors that affect portfolio return. Also, recent advances in the uncertain reasoning literature allow one to compute the marginal posterior distribution of the portfolio return even when we have a multivariate probability model with many variables. The marginal distribution of portfolio return can be dynamically updated (using Bayes rule) as we observe the values of some of the variables.

14
C. Shenoy and P.P. Shenoy
References
Chen, N., R. Roll and S. A. Ross (1986), “Economic Forces and the Stock Market,” Journal of Business, 59, 386–403. Chirinko, R. S. and G. D. Guill (1991), “A framework for assessing credit risk in depository institutions: Toward regulatory reform,” Journal of Banking and Finance, 15, 785–804. Darwiche, A. and M. Goldszmidt (1994), “On the relation between kappa calculus and probabilistic reasoning,” in R. L. Mantaras and D. Poole (eds.), Uncertainty in Artificial Intelligence: Proceedings of the Tenth Conference, 145–153, Morgan Kaufmann, San Francisco, CA. Dawid, A. P. (1979), “Conditional independence in statistical theory (with discussion),” Journal of the Royal Statistical Society, Series B, 41(1), 1–31. Dhrymes, P. J., I. Friend, and N. B. Gultekin (1984), “A Critical Reexamination of the Empirical Evidence on the Arbitrage Pricing Theory,” Journal of Finance, 39(2), 323–346. Elton, E. J., M. J. Gruber and C. R. Blake (1997), “Common factors in mutual fund returns,” Working Paper S-97-42, The NYU Salomon Center for Research in Financial Institutions and Markets, New York University, NY. Elton, E. J., and M. J. Gruber (1997), Modern Portfolio Theory and Investment Analysis, 4th Ed., John Wiley & Sons, New York. Gilks, W. R., S. Richardson, and D. J. Spiegelhalter (1996), Markov Chain Monte Carlo in Practice, Chapman & Hall, London. Goodman, L. S. (1981), “Bank Lending to Non-OPEC LDCs: Are Risks Diversifiable?,” FRBNY Quarterly Review, 10–20. Heckerman, D. (1997), “Bayesian networks for data mining,” Data Mining and Knowledge Discovery, 1, 79–119. Henrion, M. (1988), “Propagating uncertainty in Bayesian networks by probabilistic logic sampling,” in J. F. Lemmer and L. Kanal, N. (eds.), Uncertainty in Artificial Intelligence, 2, 149–164, North-Holland, Amsterdam. Henrion, M., A. Darwiche, M. Goldszmidt, G. Provan, and B. Del Favero (1994), “An experimental comparison of infinitesimal and numerical probabilities for diagnostic reasoning,” Proceedings of the Fifth Workshop on the Principles of Diagnosis, 131–139. Henrion, M., M. Pradhan, B. Del Favero, K. Huang, G. Provan and P. O'Rorke (1996), “Why is diagnosis using belief networks insensitive to imprecision in probabilities,” in E. Horvitz and F. Jensen (eds.), Uncertainty in Artificial Intelligence: Proceedings of the Twelfth Conference, 307–314, Morgan Kaufmann, San Francisco. Jensen, F. V., S. L. Lauritzen and K. G. Olesen (1990), “Bayesian updating in causal probabilistic networks by local computation,” Computational Statistics Quarterly, 4, 269–282. Jorion, P. (1997), Value at Risk: The New Benchmark for Controlling Market Risk., McGraw Hill, NY. Kao, D. (1993), “Illiquid securities: Pricing and performance measurement,” Financial Analysts Journal, 28–35. Keefer, D. L. (1994), “Certainty equivalents for three-point discrete-distribution approximations,” Management Science, 40, 760–772.

Bayesian Network Models of Portfolio Risk and Return
15
Lauritzen, S. L. and D. J. Spiegelhalter (1988), “Local computations with probabilities on graphical structures and their application to expert systems (with discussion),” Journal of Royal Statistical Society, Series B, 50(2), 157–224. Lauritzen, S. L., A. P. Dawid, B. N. Larsen and H.-G. Leimer (1990), “Independence properties of directed Markov fields,” Networks, 20(5), 491–505. Pearl, J. (1986), “Fusion, propagation and structuring in belief networks,” Artificial Intelligence, 29, 241–288. Pearl, J. (1988), Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, Morgan Kaufmann, San Mateo, CA. Pradhan, M., M. Henrion, G. Provan, B. Del Favero, and K. Huang. (1996), The sensitivity of belief networks to imprecise probabilities: An experimental investigation, Artificial Intelligence, 85(1–2), 363–397. Roll, R., and Ross, S. A. (1980), “An Empirical Investigation of the Arbitrage Pricing Theory,” Journal of Finance, 35(5), 1073-1103. Ross, S. A. (1976), “The Arbitrage Theory of Capital Asset Pricing,” Journal of economic Theory, 13, 341– 360. Shenoy, P. P. and G. Shafer (1990), “Axioms for probability and belief-function propagation,” in Shachter, R. D., T. S. Levitt, J. F. Lemmer and L. N. Kanal (eds.), Uncertainty in Artificial Intelligence, 4, 169–198, North-Holland, Amsterdam. Reprinted in: Shafer, G. and J. Pearl (eds.), Readings in Uncertain Reasoning, 575–610, 1990, Morgan Kaufmann, San Mateo, CA. Smith, J. E. (1993), “Moment methods for decision analysis,” Management Science, 39, 340–358.

对等网络模式

一、对等网简介 “对等网”也称“工作组网”，那是因为它不像企业专业网络中那样是通过域来控制的，在对等网中没有“域”，只有“工作组”，这一点要首先清楚。正因如此，我们在后面的具体网络配置中，就没有域的配置，而需配置工作组。很显然，“工作组”的概念远没有“域”那么广，所以对等网所能随的用户数也是非常有限的。在对等网络中，计算机的数量通常不会超过20台，所以对等网络相对比较简单。在对等网络中，对等网上各台计算机的有相同的功能，无主从之分，网上任意节点计算机既可以作为网络服务器，为其它计算机提供资源；也可以作为工作站，以分享其它服务器的资源；任一台计算机均可同时兼作服务器和工作站，也可只作其中之一。同时，对等网除了共享文件之外，还可以共享打印机，对等网上的打印机可被网络上的任一节点使用，如同使用本地打印机一样方便。因为对等网不需要专门的服务器来做网络支持，也不需要其他组件来提高网络的性能，因而对等网络的价格相对要便宜很多。 对等网主要有如下特点： （1）网络用户较少，一般在20台计算机以内，适合人员少，应用网络较多的中小企业； （2）网络用户都处于同一区域中； （3）对于网络来说，网络安全不是最重要的问题。 它的主要优点有：网络成本低、网络配置和维护简单。 它的缺点也相当明显的，主要有：网络性能较低、数据保密性差、文件管理分散、计算机资源占用大。 二、对等网结构 虽然对等网结构比较简单，但根据具体的应用环境和需求，对等网也因其规模和传输介质类型的不同，其实现的方式也有多种，下面分别介绍： 1、两台机的对等网 这种对等网的组建方式比较多，在传输介质方面既可以采用双绞线，也可以使用同轴电缆，还可采用串、并行电缆。所需网络设备只需相应的网线或电缆和网卡，如果采用串、并行电缆还可省去网卡的投资，直接用串、并行电缆连接两台机即可，显然这是一种最廉价的对等网组建方式。这种方式中的“串／并行电缆”俗称“零调制解调器”，所以这种方式也称为“远程通信”领域。但这种采用串、并行电缆连接的网络的传输速率非常低，并且串、并行电缆制作比较麻烦，在网卡如此便宜的今天这种对等网连接方式比较少用。 2、三台机的对等网

五款最好的免费电脑资料同步备份软件

文件夹同步就是将两个文件夹内的文件内容进行分析，可选择性的让两个文件夹内容保存一直。文件夹同步软件相当有用，虽然大多数人没用过，但它确实能够为你节省很多时间和操作。比如说：同步U盘上的数据和软件设置，查找软件版本区别和更新，同步FTP上的数据。我认为，很多情况下使用同步软件可以极大提高计算机操作效率。 高效文件同步工具GoodSync 在多种驱动设备之间自动同步和备份，（个人电脑、移动设备、网络设备）支持任何文件类型，支持多任务、多语言。人性化的界面，可自由选择部分单向双向同步，有强大的过滤系统，有完整的日志记录及更改内容报表。 注意：GoodSync分析之后会在任务文件夹生成“_gsdata_”的隐藏文件夹，里面存放在任务日志和备份文件。GoodSync有免费版和专业版之分。免费版在30天内没有任何限制，仅仅是不能可用于商业用途和政府机构。过来三十天依然可以免费使用，但是仅支持3个任务（相比很多单任务的还是强大不少）和每次100文件夹的同步工作（一般情况下够）。下载 开源同步软件FreeFileSync 界面简洁，操作简单。虽然是单任务，但是可以保存和加载配置。最重要的是，作为一款开源如软件，它没有任何限制。下载

多文件夹同步器Allway Sync Allway Sync 是一个非常容易使用的 Windows 文件同步软件。同样支持在多种设备进行同步、多向同步（1个文件夹到N个）、自动同步。有极其强大的过滤规则、错误管理，可以压缩备份、加密备份。可导出导入xml格式配置文件和任务。免费版有文件大小和数量限制。当然，有着强大功能的同时，体积和资源占用也偏大。下载

汇博通文档借阅管理组织系统软件使用使用说明

汇博通文档借阅管理系统使用说明书 汇博通知识管理系统的属性管理，实际上已提供了借阅与归还功能，但那是针对每一份文件 或档案而言的。 这里，为客户提供一款专门用于文档的借阅与归还的软件，不但可办理一份文件的借阅或归 还手续，只要有需要，也可批量办理借阅与归还，另外，还提供了与借阅有关的一系列统计 报表。 发放功能与借阅类似，所不同的只是发放不必归还，如将购买的资料、图书发放给职员学习 等。 注：借阅与归还模块的操作，需要获得以下三种权限中的一种： 系统管理员 归档授权（档案管理员） 编号授权（文件管理员） 与借阅与归还模块相关的系统参数的设置说明如下：

首页 汇博通主页的模块工具条上，有一个借阅与归还的按钮，单击它即进入借阅与归还首页。 借阅（发放） 前面已经介绍过，借阅与发放的区别在于，借阅需要归还，发放则不必归还，从某种意义上 来说，发放实际上已将所有权（或有条件的所有权）转移给接收者。 借阅界面包括左右两个子窗体，左侧子窗体用于显示可供借阅（发放）的文档，其上部有搜 索关键词输入框，输入相应关键词即可查找出可供借阅的相应文档，如果要借阅的文档已经 在操作者手上，并且，标注有条形码或电子标签，操作者可直接通过条码阅读器或电子标签

阅读器读取相应编码直接获取到该文档。 根据实际需要，通过点选左侧的复选框，选择具体文档，然后，通过点击两个子窗体中间的箭头，即可将选中的文档添加到右侧子窗体的列表中，即可直接办理借阅或发放手续。 可供借阅（发放）检索列表待选区。借阅（发放）选择勾选列表区。 可供借阅（发放） 输入文件名称、编号、责任者或主题词等属性，点击【检索】按钮进行查找，如下图： 勾选确定后点击该按钮，即可添加到已 选择列表区中。

如何使用群晖备份、同步文件

如何使用群晖备份、同步文件？通过群晖管家安装好NAS之后，想要实现备份、同步还要随时随地查看所有的文件？只需要一个Drive，就能把你的需求统统搞定。让你轻松的掌握文件同步和备份。 Drive既是备份盘、同步盘、网盘，还可以是协作盘。集中管理所有文件，还能够同步不同电脑上的数据。团队脑风暴时，可以多人在同一个文档上实时协同编辑，还能够备份电脑上的文件并且提供多版本保护。 安装及设置drive套件 1、打开群晖DSM界面，在套件中心安装Drive套件。 2、安装Drive套件会一并安装Drive管理控制台——顾名思义，就是可以设置Drive 相关功能、管理所有备份和同步的设备、查看历史版本等。 3、建议你在Drive管理控制台启用深度搜索，就可以在Drive里面查找内文关键字还有照片各种原始信息，步骤如图。

设置备份盘 1、首先进入群晖官网的下载中心，根据NAS机型，选择下载“Drive Client”PC客户端，Windows、Mac、Ubuntu一应俱全，系统兼容妥妥的。 2、PC客户端安装完成后，根据需求修改Drive服务器（NAS端）和电脑（本地端）的不同文件夹。 3、设置完成后，进入Drive PC客户端的控制面板，将同步模式改成“单向上传”，点击应用，然后就开始备份啦。 同步盘如何实现 1、同步盘很简单，设置步骤跟上面的备份盘一样，在需要同步的电脑上安装Drive PC客户端，并且选择双向同步。 2、如果在办公场景，希望把他人分享给你的共享文件夹同步到电脑本地，在PC客户 端控制面板点击“创建>启用同步与我共享”，这么一来，别人与你分享的文件也会同步到本地。

人力资源管理系统软件操作手册

XX集团—人力资源管理系统操作手册 目录 常用操作（新人必读） (2) 1.基础数据管理 ................................................................................................................... - 5 - 1.1组织架构 (5) 1.2职位体系 (8) 1.3职员维护 (11) 1.4结束初始化.................................................................................. 错误！未定义书签。 2.组织管理业务 ................................................................................................................. - 27 - 2.1组织规划 (27) 2.2人力规划 (33) 2.3组织报表 (38) 3.员工管理业务 ................................................................................................................. - 41 - 3.1员工状态管理 (41) 3.2合同管理 (41) 3.3后备人才管理 .............................................................................. 错误！未定义书签。 3.4人事事务 (52) 3.5人事报表 (59) 4.薪酬管理 ......................................................................................................................... - 69 - 4.1基础数据准备 (69) 4.2薪酬管理日常业务 (92) 4.3薪酬管理期末业务 (107) 4.4薪酬报表 (108)

对等网络配置及网络资源共享

物联网技术与应用 对等网络配置及网络资源共享 实验报告 组员：

1.实验目的 （1）了解对等网络基本配置中包含的协议，服务和基本参数 （2）了解所在系统网络组件的安装和卸载方法 （3）学习所在系统共享目录的设置和使用方法 （4）学习安装远程打印机的方法 2.实验环境 Window8，局域网 3.实验内容 （1）查看所在机器的主机名称和网络参数，了解网络基本配置中包含的协议，服务和基本参数 （2）网络组件的安装和卸载方法 （3）设置和停止共享目录 （4）安装网络打印机 4.实验步骤 首先建立局域网络，使网络内有两台电脑 （1）“我的电脑”→“属性”，查看主机名，得知两台计算机主机名为“idea-pc”和“迦尴专属”。 打开运行输入cmd，进入窗口输入ipconfig得到相关网络参数。局域网使用的是无线局域网。 （2）网络组件的安装和卸载方法：“网络和共享中心”→“本地连接”→“属

性”即可看到网络组件，可看其描述或卸载。 “控制面板”→“卸载程序”→“启用和关闭windows功能”，找到internet 信息服务，即可启用或关闭网络功能。 （3）设置和停止共享目录（由于windows版本升高，加强了安全措施和各种权

限，所以操作增加很多） 使用电脑“idea-pc”。“打开网络和共享中心”→“更改高级选项设置”。将专用网络，来宾或公用，所有网络中均选择启用文件夹共享选项，最下面的密码保护项选择关闭，以方便实验。 分享文件夹“第一小组实验八”，“右键文件夹属性”→“共享”→“共享”，选择四个中的一个并添加，此处选择everyone，即所有局域网内人均可以共享。

销售管理软件操作手册

前言 本《操作手册》内容是按该软件主界面上第一横排从左至右的顺序对各个功能加以介绍的，建议初学者先对第一章系统设置作初步了解，从第二章基础资料读起，回头再读第一章。该管理软件的重点与难点是第二章，望读者详读。 第一章系统设置 打开此管理软件，在主界面上的左上方第一栏就是【系统设置】，如下图所示： 点击【系统设置】，在系统设置下方会显示【系统设置】的内容，包括操作员管理、数据初始化、修改我的登录密码、切换用户、选项设置、单据报表设置、导入数据、数据库备份、数据库恢复、压缩和修复数据库、退出程序。下面分别将这些功能作简要介绍： 1.1操作员管理 新建、删除使用本软件的操作员，授权他们可以使用哪些功能。此功能只有系统管理员可以使用。 1.1.1 进入界面 单击【系统设置】，选择其中的【操作员管理】，画面如下：

1.1.2、增加操作员 单击【新建】按钮，画面如下： 输入用户名称、初始密码、选择用户权限，可对用户进行适当描述，按【保存】后就点【退出】，就完成了新操作员的添加，效果如下图。

1.1.3 删除操作员 选择要删除的操作员，单击【删除】按钮。 1.1.4 修改操作员 选择要修改的操作员，单击【修改】按钮，可对操作员作相应修改，修改后需保存。 1.1.5 用户操作权限 选择要修改的操作员，单击【修改】按钮，出现以下画面，点击【用户权限】栏下的编辑框，出现对号后点【保存】，该操作员就有了此权限。 1.2数据初始化 1.2.1进入界面 单击【系统设置】，选择其中的【数据初始化】，画面如下：

1.2.2数据清除 选择要清除的数据，即数据前出现对号，按【确定】后点【退出】，就可清除相应数据。 1.3 修改我的登录密码 1.3.1进入界面 单击【系统设置】，选择其中的【修改我的登录密码】，画面如下： 1.3.2密码修改 输入原密码、现密码，然后对新密码进行验证，按【确定】后关闭此窗口，就可完成密码修改。 1.4 切换用户 1.4.1进入界面 单击【系统设置】，选择其中的【切换用户】，画面如下：

对等网络(P2P)总结整理解析

对等网络(P2P 一、概述 (一定义 对等网络(P2P网络是分布式系统和计算机网络相结合的产物,在应用领域和学术界获得了广泛的重视和成功,被称为“改变Internet的新一代网络技术”。 对等网络(P2P:Peer to Peer。peer指网络结点在: 1行为上是自由的—任意加入、退出,不受其它结点限制,匿名; 2功能上是平等的—不管实际能力的差异; 3连接上是互联的—直接/间接,任两结点可建立逻辑链接,对应物理网上的一条IP路径。 (二P2P网络的优势 1、充分利用网络带宽 P2P不通过服务器进行信息交换,无服务器瓶颈,无单点失效,充分利用网络带宽,如BT下载多个文件,可接近实际最大带宽,HTTP及FTP很少有这样的效果 2、提高网络工作效率 结构化P2P有严格拓扑结构,基于DHT,将网络结点、数据对象高效均匀地映射到覆盖网中,路由效率高 3、开发了每个网络结点的潜力 结点资源是指计算能力及存储容量,个人计算机并非永久联网,是临时性的动态结点,称为“网络边缘结点”。P2P使内容“位于中心”转变为“位于边缘”,计算模式由“服务器集中计算”转变为“分布式协同计算”。

4、具有高可扩展性(scalability 当网络结点总数增加时,可进行可扩展性衡量。P2P网络中,结点间分摊通信开销,无需增加设备,路由跳数增量小。 5、良好的容错性 主要体现在:冗余方法、周期性检测、结点自适应状态维护。 二、第一代混合式P2P网络 (一主要代表 混合式P2P网络,它是C/S和P2P两种模式的混合;有两个主要代表: 1、Napster——P2P网络的先驱 2、BitTorrent——分片优化的新一代混合式P2P网络 (二第一代P2P网络的特点 1、拓扑结构 1混合式(C/S+P2P 2星型拓扑结构,以服务器为核心 2、查询与路由 1用户向服务器发出查询请求,服务器返回文件索引 2用户根据索引与其它用户进行数据传输 3路由跳数为O(1,即常数跳 3、容错性:取决于服务器的故障概率(实际网络中,由于成本原因,可用性较低。

多文件夹的自动同步和各向同步工具

多文件夹的自动同步和各向同步工具 出处：小建の软件园作者：佚名日期：2008-06-25 关键字：同步 对于经常需要备份文件，同步文件的网友，Allway Sync 可谓不可多得，虽然不能激活其专业版，对文件数量多和经常性的同步操作可能会超过免费版的限制，不过对于一般文件数量不多同步操作可以完全满足，Allway Sync 使用相当简单，多种同步方式能满足你不同需求。对重要文件进行备份是文件恢复最好的方法，而 Allway Sync 可以简化你许多备份的过程，能实现自动备份，如果你“胃口”不大，免费版应当已经可以满足。 下载地址：https://www.wendangku.net/doc/af3807460.html,/soft/23495.html Allway Sync 可以进行自动同步，可以对的文件/文件夹进行筛选，只备份需要的东东。

Allway Sync 备份方式介绍 - 同步方式有源文件夹同步和各向同步两种方式： 1、源文件夹同步方式将以一个文件夹为基准，删除或覆盖其余文件夹与源文件相比较不相同的文件。 2、各向同步方式则自动将更新的文件覆盖几个同步文件夹中的旧文件。软件带有一个小型数据库，监视每次更新后的文件状态。如果在一次同步之后，你删除了同步文件夹中某些文件，它在同步的时候将其它的几个文件夹的副本也删除，而不会将不需要的未删除文件重复拷贝到已更新的文件夹。由于软件自己会对文件进行删除和覆盖，它提供了使用回收站进行文件备份的措施，使用者可以在不慎执行错误的同步动作之后，从回收站将错误删除或覆盖的文件找回来（默认禁用该功能，请到软件选项处激活相应设置）。 主程序在 AllwaySync＼Bin＼里面，此为多国语言版，在语音选项那里选择中文即可。不过退出的时候会有错误提示（貌似没影响？）

对等网络的网络弹性分析

对等网络的网络弹性分析 摘要：网络弹性研究的是网络在节点失效或被有意攻击下所表现出来的特征。分析Gnutella网络的网络弹性，包括对于随机攻击的容错性和对于选择性攻击的抗攻击性，并与ER模型和EBA模型进行了对比。Gnutella网络对于随机攻击具有很好的容错性，但是对于选择性攻击却显得脆弱。最后对网络弹性进行了理论分析，给出了网络在出现最大集团临界点之前的平均集团大小的公式解。 关键词：对等网络；无标度；网络弹性；脆弱性 中图分类号：TP393.02文献标识码：A 文章编号：1001-9081(2007)04-0784-04 0 引言 在过去的40多年里，科学家习惯于将所有复杂网络看作是随机网络。随机网络中绝大部分节点的连结数目会大致相同。1998年开展的一个描绘互联网的项目却揭示了令人惊诧的事实：基本上，互联网是由少数高连结性的页面串联起来的，80%以上页面的连结数不到4个，而只占节点总数不到万分之一的极少数节点，例如门户网Yahoo和搜索引擎Google等类似网站，却高达上百万乃至几十亿个链接。研究者把包含这种重要集散节点的网络称为无标度网络[1]。

具有集散节点和集群结构的无标度网络，对意外故障具有极强的承受能力，但面对蓄意的攻击和破坏却不堪一击[2]。在随机网络中，如果大部分节点发生瘫痪，将不可避免地导致网络的分裂。无标度网络的模拟结果则展现了全然不同的情况，随意选择高达80%的节点使之失效，剩余的网络还可能组成一个完整的集群并保持任意两点间的连接，但是只要5%―10%的集散节点同时失效，就可导致互联网溃散成孤立无援的小群路由器。 许多复杂网络系统显示出惊人的容错特性，例如复杂通信网络也常常显示出很强的健壮性，一些关键单元的局部失效很少会导致全局信息传送的损失。但并不是所有的网络都具有这样的容错特性，只有那些异构连接的网络，即无标度网络才有这种特性，这样的网络包括WWW、因特网、社会网络等。虽然无标度网络具有很强的容错性，但是对于那些有意攻击，无标度网络却非常脆弱。容错性和抗攻击性是通信网络的基本属性，可以用这两种属性来概括网络弹性。 对等网络技术和复杂网络理论的进展促使对现有对等 网络的拓扑结构进行深入分析。对网络弹性的认识可以使从网络拓扑的角度了解网络的脆弱点，以及如何设计有效的策略保护、减小攻击带来的危害。本文研究Gnutella网络的网络弹性，并与ER模型和EBA模型进行了比较，对比不同类 型的复杂网络在攻击中的网络弹性。当网络受到攻击达到某

备份软件使用方法v1.0

备份软件使用方法 一Bestsync2012使用说明 1 软件运行 点击BestSync2012运行软件 2 设置任务 在编辑菜单下点击追加任务（如果任务列表下没有任务可以在文件菜单下选择新建任务选项） 软件会弹出任务窗口，用来设置同步任务

以其中一个任务为例

选择好同步的文件夹和同步方向，点击下一步，按照要求设置任务即可。 3 查看任务 在以有任务中点击设置任务（任务必须是未在同步状态，否者不能点击设置任务选项）

点击后软件会弹出设置同步任务窗口，在这里可以在里面进行任务修改和设置

目前我们设置的同步任务只需要修改一般和日程两个窗口下的内容，其他暂时不需要修改。 BestSync2012这款同步软件目前还不是很稳定，需要不定期检查一下软件是否运行正常，如果发现软件出错，就关闭软件后在打开BestSync2012软件，因为打开软件后软件不会自动启动同步功能，所有需要手动启动所有任务 注意： 1 在修改任务在开启后，必须将修改的任务停止一下在开启，不然同步任务不能正常同步。 2 现有BestSync2012同步软件在16.15和151.247这两台机器上。

二Backup Exec 2010 R2 SP1使用说明 1 软件运行 点击Backup Exec 2010运行软件 2 设置任务 在作业设置选项中可以看到作业的作业名称、策略名称和备份选这项列表。 其中作业名称里放有现有作业，双击其中一个作业就可以看到作业属性。作业属性默认显示设备和介质窗口，在设备和介质窗口下可以选择设备和介质集。目前设备选项中因为只有一台磁带机工作，所有只有一个选项，而介质集一般选择永久保留数据-不允许覆盖选项。

管理软件使用说明书

目录 1 软件介绍...................................................... １ 2 软件运行环境 ................................................. １ 3 软件安装步骤 ................................................. １ 4 软件卸载步骤 ................................................. ４ 5 软件使用...................................................... ４5.1、创建数据库.............................................................................................................................. ４ 5.2、创建数据数据表................................................................................................................... ６ 5.3、历史数据读取 ........................................................................................................................ ７ 5.4、查看历史数据、通道信息.............................................................................................. ８ 5.5、打印数据、曲线或图片输出 .................................................................................... １３ 5.6、数据实时采集 .................................................................................................................... １５ 6 软件使用中可能出现的问题与解决方法.................. １８6.1、不出现对话框 .................................................................................................................... １８ 6.2、数据库不能建立............................................................................................................... １８ 6.3、U盘不能数据转存........................................................................................................... １８ 6.4、U盘上没有文件 ................................................................................................................ １８ 6.5、U盘数据不能导入计算机；...................................................................................... １８

GoodSync同步软件完美注册

GoodSync同步软件完美注册、本地同步图文教程 出处：西西整理作者：西西日期：2012-4-12 15:22:15 [大中小] 评论: 0 | 我要发表看法 文件管理这件看似简单的事，真的不简单，因为为了防止意外情况，你需要对文件进行备份，时间一久随着文件数量的增加，再加上有时也会临时队备份文件进行修改等。再想查出这个是最新的、文件有木有全部备份等….就没那么容易了吧！其实这一切说了很简单，因为你可以请：GoodSync软件来帮忙！ GoodSync是一款简单可靠的文件备份和文件同步软件，可以实现两台电脑或者电脑与U盘之间的数据文件的自动同步。GoodSync可以在本地U盘与电脑之间，以及U盘、移动硬盘或电脑与服务器、外部驱动器、W indowsM obile设备、网友、网盘等之间自动同步或单向备份数据。它能自动分析、同步、备份您的电子邮件、珍贵照片、联系人、电影视频、音乐文件、财务文件和其它重要文件。再也不会遗失您的电子邮件，照片，MP3等。 由于GoodSync为共享收费软件，所以这次西西带来的是官网原版+注册机（下载地址，下载的压缩包内含官网下载的GoodSync v9.1.5.5主程序和注册机以及注册说明），还是那句老话：如果你有能力请支持购买正版的GoodSync，如果….就低调吧！好吧！一起来看下注册方法吧！ GoodSync 注册方法： 1、首先下载压缩包，并解压运行GoodSync-Setup.exe 进行软件安装，软件默认安装为英文，如果要安装简体中文版，在安装时注意选择语言为：simpchinese项，安装完毕后运行GoodSync程序。 2、将你电脑的系统时间设置到2011年。 3、如下图所示，在软件主界面依次点击选择：帮助→ 激活专业版。

东莞二期投标文件管理软件操作手册V2.0.0.3

投标文件管理软件（V2.0.0.3） 用 户 使 用 手 册 深圳市斯维尓科技有限公司 二〇一三年三月五日

目录 1引言 (3) 2 程序运行环境 (4) 3 程序安装 (4) 4 软件启动 (9) 5软件整体说明 (12) 6 软件操作说明 (15) 6.1导入查看招标文件 (15) 6.2新建投标文件 (16) 6.3投标文件的管理功能 (23) 6.4校对工程量清单 (29) 6.5转换投标文件 (30) 6.6 电子签章 (32) 6.7生成投标文件 (34) 6.8查看数字签名信息 (41) 7 程序卸载 (42)

1引言 编写本手册的主要目的是为东莞市建设工程交易中心电子评标系统的投标文件管理软件的使用提供帮助。 投标文件管理软件主要提供给投标单位使用。投标单位通过投标文件管理软件将工程招标文件的一些主要内容导出，根据招标要求制作投标文件；加入已经制作好的工程投标文件所包含的所有文档（包括：技术标文件、工程量清单、工程图纸以及其它文件等），并进行管理，对文件包进行CA数字签名以防篡改，并生成压缩加密的电子投标文件包的功能。 投标文件管理软件的使用总体流程如下图所示：

2 程序运行环境 ?硬件环境：CPU: P4 2GHZ 内存2G,硬盘80GB ?软件环境：Windows 2000/XP/Windows Server 2003 ?软件支持：OFFICE2007+PDF转换插件/OFFICE2010 ?网络环境：带宽10/100Mbps 3 程序安装 东莞市建设工程交易中心网站(https://www.wendangku.net/doc/af3807460.html,/)上下载最新安装包，点击安装程序，安装程序引导用户进行系统安装，主要有以下步骤： 一、启动安装程序，进入安装系统欢迎界面。如下图：

建立对等网详细步骤

实验一建立对等网 一、实验目的 (1)熟悉10BASE-T星型拓扑以太网的网卡、线缆、连接器等网络硬件设备; (2)熟悉WINDOWS中的网络组件及各参数的设置; (3)理解对等网络的特点。 二、实验环境 此实验的基本要求就是两台以上计算机作为一个工作组,连接到一台服务器上,建立一个基于Windows的对等网络,物理结构为10BASE-T以太网。各工作组中的用户可以共享资源。 三、实验内容 (1)网络布线 EIA/TIA的布线标准中规定了两种双绞线的线序568A与568B,分别为: T568A:白绿 | 绿 | 白橙 | 蓝 | 白蓝 | 橙 | 白棕 | 棕 T568B:白橙 | 橙 | 白绿 | 蓝 | 白蓝 | 绿 | 白棕 | 棕 在整个网络布线中应用一种布线方式,但两端都有RJ45端头的网络连线无论就是采用端接方式A,还就是端接方式B,在网络中都就是通用的。实际应用中,大多数都使用T568B的标准,通常认为该标准对电磁干扰的屏蔽更好,本次实习中即采用了端接方式B。 (2)连接网线,建立对等网 连接网线的方式与网卡接口与网络结构有关,本局域网中采用的就是星型结构。以集线器(HUB)为中央结点,网络中所有计算机都通过双绞线连接至集线器,通过集线器交换信息。星型结构的优点就是利用中央结点可方便地提供服务与重新配置网络,单个连接点的故障只影响一个设备;缺点就是每个站点直接与中央结点相连,需要大量电缆,费用较高。 连接好网络线后接通计算机电源,观察网卡后面板上的两只LED工作状态指示灯。绿灯亮表示网络线接通,红灯间接闪烁说明网卡工作正常。 (3)MS-DOS方式中,执行ping命令进行测试

个人文件同步备份FILEGEE

软件简介： FileGee之软件主界面(图一) FileGee个人文件同步备份系统是一款优秀的文件同步与备份软件。它集文件备份、同步、加密、分割于一身。协助个人用户实现硬盘之间，硬盘与移动存储设备之间的备份与同步。强大的容错功能和详尽的日志、进度显示，更保证了备份、同步的可靠性。高效稳定、占用资源少的特点，充分满足了用户的需求。不需要额外的硬件资源，便能搭建起一个功能强大、高效稳定的全自动备份环境，是一种性价比极高的选择。 一.软件安装 FileGee个人文件同步备份系统在使用前须对其进行安装才可进行使用，软件须按照提示进行安装，软件安装过程如下图所示： FileGee之接受安装协议(图二) FileGee之选择安装目录(图三) FileGee之完成安装(图四) 二.软件使用 FileGee在完成安装后双击桌面图标即可启动该软件，用户如要创建备份，即可点击软件左上角处的新建任务按钮来创建新任务，软件提供多种任务类型，如单向同步，双向同步，镜像同步，更新同步等等，用户鼠标停留在任务类型上即可看到相关的解释说明，如下图所示： FileGee之新建任务(图五) 用户在选择创建备份任务类型后，即可点击下一步按钮，点击后软件会自动弹出窗口，用户需在窗口中设置要进行备份的文件夹所在位置，如下图所示： FileGee之设置备份文件夹(图六) 设置完要进行备份的文件夹后，我们还需要对备份文件的存储位置进行设置保存，另外为了节省空间我们还可以对文件设置是否进行压缩，如下图所示：

FileGee之备份文件保存(图七) 在设置完毕后我们即可点击下一步按钮，在后面的设置选项中我们还可以对备份的文件进行详细设置，如是否包含源目录的子目录，还可以根据文件名对要备份的文件进行过滤，也可以对文件进行过滤设置，如下图所示： FileGee之备份设置(图八) 设置完毕后，我们即可点击软件上侧列表中的开始按钮对文件进行备份，，另外还可以点击软件上侧的定时自动功能设置定时对文件夹进行自动备份，如下图所示： FileGee之备份任务(图九) 小结：FileGee作为一款免费得文件夹自动同步备份工具，不但功能上比较强大，在使用上也是非常的方便，如果您也需要一款文件备份工具的话，那么就来试试FileGee吧，只需简单几步就可以完成文件夹同步备份，非常方便!

说明书金助手美容美发管理软件操作手册连锁

说明书金助手美容美发管理软件操作手册连锁 Document number【AA80KGB-AA98YT-AAT8CB-2A6UT-A18GG】

金助手美容美发管理软件操作手册（连锁版） 后台设置 打开金助手管理系统进入主界面 操作员常用的功能都放在了主界面上，下面介绍下如何设置后台参数。打开主界面的右上角的按键，

一、基本参数设置 打开基本参数设置 在基本参数设置中进行本店卡种设置、基本卡种设置、基本折扣标准设置、基本工种设置和当前活动（现金）结帐执行的折扣标准。 首先在基本卡种设置中添加本店发行的所有会员卡种类（例如：本店发行金卡、银卡、钻卡和会员卡）同时系统会自动生成和卡种名称相同的折扣标准。 在基本工种设置中添加本店的员工的工种（例如：美容师、美发师、美甲师、助理等） 在当前活动（现金）结帐执行的折扣标准中添加散客所享受的折扣标准。（例如：店庆时，所有散客享受和金卡会员相同的折扣） 二、设置分店信息 设置分店信息是指多店连锁的情况下，在使用本软件的时候需要首先设置不同的分店信息。 （例如：***一店、***二店、***三店等） 点击对话框右侧的新增分店来添加分店信息 分店信息设置后了以后，同时设置分店对应的库房信息包括库房名称、编号、是否设置为默认销售出货库房等。

三、公共参数设置 打开公共参数设置，包括精确度方式、服务类别设置、商品类别设置、计次项目类别设置 部门设置、记事本数据类型、其他支付方式、其他积分类型以及缺勤原因等项目的设置。

精确度方式设置 这里的功能主要体现的是结帐时出现的零碎钱 （例如：顾客做了项目后结帐，原本顾客的价位是320元，折后的金额是元，折后的金额出现了元的零钱，这时我将系统设置成→，此时结帐的金额就会显示成为264元；当然我们也可以这样设置→，此时结帐的金额就会显示成为263元。同样的道理，264元中的4元零钱也可以用上面的方式进行取舍。）服务类别设置 设置服务类别时，最重要的是设置公共类别，设置了公共类别后可以更快捷，更方便的设置员工的提成系数。 分类一和分类二是可以按照项目的用途或是品牌来划分项目分类。 此功能主要是对服务项目的类别进行划分。操作员可以根据本店的自身情况进行设置。

备份与恢复应用

备份与恢复应用 备份技术 数据备份方式 从数据备份方式来说，主要有映像备份与逐文件备份两种方式。拓普恒基NAS产品主要采用的是逐文件备份方式。 通过进入文件系统，阅读文件结构，以及从一个介质到另一个介质复制文件，从而生成新文件结构。它可针对单独文件生成备份。逐文件备份比映像备份安全，因为整个文件结构都复制了。因而允许信息迁移入不同的格式或设备类型。逐文件备份还允许用户恢复个别文件或执行部分备份。在存在变化而信息无法恢复至同类介质的情况下，逐文件备份更安全。 逐文件备份通常恢复的时间要长于备份。当需要恢复单独文件和针对大型文件，如数据库文件时，建议使用逐文件备份。 数据备份策略 NAS在实现数据备份的时候能够支持两种备份策略，用户可以根据自己的应用环境来确定选用那种备份策略，在选择的时候，了解文档位的作用十分重要。文档位是一种标志，存在于每个文件中，以表明文件已完成修改的时间。一些备份设施使用文档位以跟踪文件备份状态和其他使用日志。 我们的技术支持的两种逐文件备份方式为：全备份和增量备份。 1、全备份： 全面系统备份将把所有文件、目录、用户信息、安全属性和系统/操作系统文件复制到备份设备。当执行全面系统备份时，无需检查文档位，因为所有文件都将备份。每个备份计划都应包括全面备份。 2、增量备份 增量备份只复制上次备份后发生变化的文件。备份软件将检查文档位，以确定文件是否被修改，以及是否需要备份。如果文件的文档位表明为新文件或已修改，文件将被复制到备份设备，文档位将清除。 两种逐文件备份方式的图示如下： 备份策略的选择并非完全以围绕数据备份的问题为基础，在选择最佳策略时也必须考虑到恢复的问题。

ERP管理软件操作手册

1.软件登录 双击进入看到如下界面，如图选择自己所需的公司帐套双击进入 操作 进入之后看到如下界面，如图选择自己的用户名（有密码输入密码）按键盘的F8或界面上的确认进入 进入界面看到如下界面，如图：左边为常用的报表查询单据；右边则为 常用单 据的操作及基础资料的设置；下方及一些软件自带按钮和右下角显示公司名称和登录用户名 2.单据的通用界面及功能键介绍 如下图：第一排：单据常用功能按钮；第二排：单据表头上方操作界面；第三排：单据表身操作界面；第四排：单据表头下方操作界面 下图为常用功能键的功能介绍；新增：添加新单据时使用；速查:查找原有的单据使用；编辑：修改原有单据使用；删除：删除单据使用;打印：打印出来使用；存盘：单据新增、修改后保存使用 3.单据操作指导 如下图，此界面为技术部业务的常用操作单据及报表总体介绍。 增加货品资料操作。 点击进入界面，如图操作。

查询原有货品资料操作。 查询结果 查询条件多种，根据实际方便来操作。 4.关于虚拟货品的替代件的输入操作。 点击进去 基本操作如查询跟货品操作一样，不一一讲解。 两种替代方式介绍 1.补量替换： 2.全量替换： 替代比例； 5.建立标准成品BOM的操作. 点击进入界面如下图。 增加BOM操作跟货品操作步骤一致，如图： 当BOM确定后，还未产生后续操作时，需修改BOM。如下操作。

当单据已经产生后续操作时，要修改该BOM则进入 进入操作。 三种活动方式介绍 1.增加：在原有BOM 里增加新的物料； 2.删除: 删除原有BOM里的物料； 3.改变：改变原有BOM里的物料货品的数量等； 6.客户订制品BOM的维护操作。 点击进入界面进行操作，如下图： 当单据已经产生后续操作时，要修改该订单配方则进入进入 操作，操作方法跟一样，就不一一细讲。

网络同步备份镜像备份软件使用分享

SyncBackPro网络同步备份软件教程 单位：华兴科软-技术服务部部门经理：余海教材开发：李江涛 不管你是不是一名网络技术运维工程师，你一定想过想把你家里的电脑与办公室的电脑文件能够保持同步，不要每天带着U盘把文件拖来拖去，不管你是不是想让你的个人电脑与办公室电脑的文件能够实时同步，你因为也许误删了一些文件，无法恢复想把自己的手剁下来喂狗。不管怎么样你总是会遇到这样那样的情况需要同步或者备份你当前电脑中的文件或者资料，但是百度网盘取消了文件夹同步功能，360网盘上传速度让人捉急。Linux上干脆连以上两款网盘备份软件都没有，可怜的你遇到问题只能默默地躲在厕所里哭泣。 今天本教程就是要拯救你，拯救受苦受难的大众，通过SyncBackPro你将能达成所愿，不再后悔，不再哭泣，让你不花钱也能实现普通人的容灾备份，长话短说，下面我们正式开始。 首先，你当现使用的必须是一台win7及以上版本的电脑，然后你在下面网址处 http://www.dayanzai.me/syncbackpro.html下载并且按照说明安装该软件，当然最好使用正版，这样才能获得长期稳定的更新和维护。 按照以上要求完成软件安装后，你在开始菜单和左面中都找不到软件的快捷方式——没关系！，点击开始菜单在所用程序/所有应用中找到2BrightSparks这个文件夹，点击进去，就能看到软件的快捷方式，当然你可以把它发送到桌面上，全凭你个人的意愿。 图一

图二 通过快捷方式打开软件会看到如下界面 好吧，我承认这个界面确实有点单调，不过一会儿你就不会这么觉得了，你可以在上方的菜单点击【同步任务】—【添加】来添加你第一项备份或者同步任务也可以点击下方的快捷菜单中的【添加】来操作。点击【添加】后如下图所示：你需要输入你的任务名称，这个根据你实际的需求来写，比如：文件备份。

备份与恢复管理相关的安全管理制度

信息系统备份与恢复管理 第一章总则 第一条为保障公司信息系统的安全，使得在计算机系统失效或数据丢失时，能依靠备份尽快地恢复系统和数据，保护关键应用和数据的安全，保证数据不丢失，特制定本办法。 第二条对于信息系统涉及到的网络设备、网络线路、加密设备、计算机设备、应用系统、数据库、维护人员，采取备份措施，确保在需要时有备用资源可供调配和恢复。 第三条本管理办法中涉及到的设备主要指运行在信息技术部主机房中的网络设备、加密设备及计算机设备。

第四条信息系统备份手段根据不同信息的重要程度及恢复时间要求分为实时热备份和冷备份等。同一平台的系统应尽量使用同样的备份手段，便于管理和使用。信息技术部负责信息系统的备份与恢复管理，并制定数据备份计划，对数据备份的时间、内容、级别、人员、保管期限、异地存取和销毁手续等进行明确规定。第五条信息技术部应根据各系统的重要程度、恢复要求及有关规定要求制定系统配置、操作系统、各应用系统及数据库和数据文件的备份周期和保存期限。 第六条对于重要系统和数据的备份周期及备份保存期限应遵循以下原则： (一) 至少要保留一份全系统备份。 (二) 每日运行中发生变更的文件，都应进行备份。

(三) 生产系统程序库要定期做备份，每月至少做一次。 (四) 生产系统有变更时，须对变更前后的程序库进行备份。 (五) 批加工若有对主文件的更新操作，则应进行批加工前备份。 (六) 每天批加工结束后都要对数据文件进行批后备份，对核心数据须进行第二备份。 (七) 对批加工生成的报表也要有相应的备份手段，并按规定的保留期限进行保留。 (八) 用于制作给用户数据盘的文件应有备份。 (九) 各重要业务系统的月末、半年末、年末以及计息日等特殊日的数据备份须永久保留。 (十) 定期将生产系统的数据进行删减压缩，并将删减的数据备份上磁带，永久保留。 (十一) 以上未明确保存期限的各项备份的保存至少应保存一周。